THE ROLE OF COMPETITION IN EFFECTIVE OUTSOURCING: SUBSIDIZED FOOD DISTRIBUTION IN INDONESIA

Gratis

0
0
72
9 months ago
Preview
Full text

  PAPER 1 - 2018

WORKING THE ROLE OF COMPETITION

IN EFFECTIVE OUTSOURCING:

SUBSIDIZED FOOD DISTRIBUTION

  (MIT) Abhijit Banerjee

THE ROLE OF COMPETITION

IN EFFECTIVE OUTSOURCING:

  

Abhijit Banerjee (MIT), Rema Hanna (Harvard University),

Jordan Kyle (IFPRI), Benjamin A. Olken (MIT),

Sudarno Sumarto (TNP2K and SMERU)

  

TNP2K WORKING PAPER 1 - 2018

February 2018

The TNP2K Working Paper Series disseminates the findings of work in progress to encourage

discussion and exchange of ideas on poverty, social protection and development issues.

  

Support to this publication is provided by the Australian Government through the MAHKOTA

  Effective Outsourcing:

The Role of Competition in Effective Outsourcing:

  1 Subsidized Food Distribution in Indonesia

Abhijit Banerjee, MIT

  

Abhijit Banerjee, MIT

Rema Hanna, Harvard University

  

Rema Hanna, Harvard University

Jordan Kyle, IFPRI

  

Jordan Kyle, IFPRI

Benjamin A. Olken, MIT

  

Benjamin A. Olken, MIT

Sudarno Sumarto, TNP2K and SMERU

  

Sudarno Sumarto, TNP2K and SMERU

March 2017

  

ABSTRACT

Abstract

  S hould government service delivery be outsourced to the private sector? If so, how? We conduct the first randomized field experiment on these issues, spread across 572 Indonesian localities. We show that allowing for outsourcing the last mile of a subsidized food delivery program reduced operating costs without sacrificing quality. However, prices paid by citizens were lower only where we exogenously increased competition in the bidding process. Corrupt elites attempted to block reform, but high rents in these areas also increased entry, offsetting this effect. The results suggest that sufficient

  Table of Contents

  I. Introduction

  II. Setting, Experimental Design, and Data

  III. Contracting Out and the Level of Competition

  IV. The Role of Existing Rents: The Race Between Vested Interests and Entry

  V. Conclusion Works Cited Appendix

  1-6 6-14 14-19 19-22 22-23 24-26 27-64

I. INTRODUCTION

  Should the state directly provide public services, or should it contract out service delivery to a private provider? In the seminal paper by Hart, Shleifer, and Vishny (1997, HSV henceforth), private contractors may potentially deliver better services—i.e. higher quality and/or a lower price—because governments can provide stronger incentives to private contractors than to their own employees. These s tronger incentives may potentially lead to efficiency improvements, but come with a risk that a contractor may lower quality below socially efficient levels to cut their costs. Therefore, in settings where non-contractible quality dimensions may be important—prisons are HSV’s example—public provision may be preferred.

  Even if contracting out can potentially improve efficiency, there is no guarantee that the public will benefit for two distinct reasons. First, if there is limited competition for the contract, the gains may accrue entirely to the contractor rather than the public. While enacting regulations to ensure sufficient competition is the standard remedy to this problem, it is possible that competitive bidding processes 2 can also exacerbate the quality problems discussed above or lead to undercutting to win contracts. Indeed, Bajari, McMillan, and Tadelis (2009) argue that if the contracts are sufficiently complex, competitive bidding may perform worse than simply negotiating with a single firm. Moreover, while

  3

  the standard view is that more bona fide bidders improves outcomes, the regulatory imposition of a requirement that there be a minimum number of bidders may bring in non-serious bidders and muddy the choice process.

  

mechanisms have a dynamic of their own: it is possible that the same rents that the incumbents want to

protect will attract entrants, and the rents will get competed away.

  This paper is an attempt to assess the relevance of these competing forces and the overall

benefits of outsourcing, in a context, rural Indonesia, where there is limited on the ground expertise in

how to do procurement and, as we will see, corruption is rife. To this end, we carry out a randomized

control trial of outsourcing across 572 localities in rural Indonesia. We first test whether contracting out

leads to efficiency gains and whether these gains come at a cost of lower quality. We then explore

whether the level of competitiveness of the bidding process affects the outcomes by randomly varying

the degree of competition among areas selected for outsourcing. Finally, to examine the potentially

conflicting effects of baseline rents, we examine how the outsourcing process, entry decisions, and

ultimately the outcomes for the consumers differs across areas that we associate with high and low

levels of rents.

  We study these questions in the context of the last-mile delivery of rice in Raskin, Indonesia’s

largest targeted transfer program (with an annual budget of over US$1.5 billion). Under Raskin, eligible

households receive a monthly allocation of subsidized rice. As is typical in most developing countries,

even though this is a central government program, the process of transferring rice from central

government warehouses to beneficiaries –the “last mile” – is administered locally by either the locality

  5 head or someone he designates as social welfare coordinator.

  The distribution of rice in Raskin is plagued by several challenges in this “last mile.” While the

  

distributor (e.g. who substituted inferior rice, or delayed picking it up), or if it originated in the central

government warehouse (e.g. which gave out bad rice to begin with, or was out of stock). Moreover, real

procurement challenges may exist: there may be inadequate competition for a job of this size from

people competent to do it, those administering the procurement procedures may have limited experience

or understanding of procurement procedures, and/or local leaders who get rents from status quo may

try to sabotage the process or steer the bidding process to their favored bidders.

  To examine these questions, in 191 randomly selected localities out of the 572, the central

government introduced a procedure that allowed for competitive bidding for the right to distribute

Raskin locally. Bids specified the markup to be charged, as well as other aspects of the distribution

process (e.g. where rice would be distributed, when copays would be collected). The selection rule used

to choose among bidders was not imposed externally, but instead a small, local committee examined

the bids and selected the winner. The incumbent local government distributor was also given the option

to bid, providing the committee with the option to keep the status quo. In short, this created a process

that allowed citizens to compete with the government provider for the job.

  The bidding process by its very nature increases transparency since people need information on

how the current process works in order to decide whether and how much to bid. Thus, we also randomly

assigned an additional 96 localities (out of the 572) to have the same set of meetings to describe

the current processes, but not the actual bidding. This information-only treatment serves as a placebo

comparison group that allows us to disentangle whether any observed effects are driven by allowing for

  

costs and ‘compensation’ paid by the distributors. We find no declines on other dimensions in either

bidding treatment: the quantity of rice received did not change nor did the quality of the distribution

process (e.g. quality of the rice, time to pick up rice) decline, and rice quality may have even improved.

In short, outsourcing has the potential to improve outcomes, but only with sufficient competition in

procurement does the public share in its gains.

  We then investigate whether capture by elites was an important factor limiting the impact of

the program. As the municipal head (or someone he designates) is the incumbent supplier, he may put

road

  • blocks in the contracting if he obtains substantial rents from the process. Local officials could do this ex-ante by preventing the bidding process from occurring at all or by discouraging people from

    bidding, thus limiting or eliminating competition in the process. Or, they could allow a competitive

    bidding process to occur but block the winning bidder ex-post from actually assuming responsibility for

    distribution.

  We test whether this type of blocking behavior is higher in cases when the incumbent was

enjoying higher rents from the program. While it is challenging to measure the rents directly, two pieces

of evidence suggest that higher baseline Raskin prices are consistent with larger rents, and not just with

higher transportation or other costs. First, high baseline markups are strongly correlated to households’

perceptions of the level of corruption of the municipal head and the incumbent Raskin distributor.

Second, this appears to be more than just dissatisfaction with being charged a high price. Following

Fischbacher and Follmi-Heusi (2013) and Hanna and Wang (2015), we elicit an experimental measure

  In short, giving localities the option to contract out delivery of government services works, but

only when there is sufficient competition among potential providers. This paper builds on several

literatures, notably the literature on outsourcing government services and that on the role of competition

in procurement. With respect to outsourcing, Levin and Tadelis (2010) focus on which services are

likely to be privatized and where, focusing on the tradeoffs between contracting costs, the potential

efficiency gains from outsourcing, and the potential political reasons not to privatize. Our results on

the tension between protecting rents and efficiency gains from protecting rents echo these findings.

  6 There is an extensive theoretical literature on the role of competition in procurement auctions.

  

One important point from this literature is that while there are many situations where competition

reduces rents going to the suppliers, this is by no means inevitable. For example, Bulow and Klemperer

(2002) and Hong and Shum (2002) observe that in common value auctions the presence of more bidders

can worsen the winner’s curse and lead to more defensive bidding by all participants. Compte and Jehiel

(2002) make a similar point about affiliated value auctions and more recently, Li and Zheng (2009)

show that even in first price private value auctions the presence of a higher number of potential bidders

can increase the rents going to the suppliers when there is endogenous costly entry because a higher

number of potential bidders can reduce the probability that any given bidder will enter by so much that

the total number of actual bidders goes down.

  It is plausible that our setting has both private values and common values. Each potential

supplier knows more about his or her own cost of time and ability to transport rice than the rest of the

population. On the other hand, the incumbent probably knows more about just how much work it is to

  By contrast, Hong and Shum (2002) and Li and Zheng (2009), who also empirically examine

the effect of competition in an auction setting, use the observational variation in competition across

different auctions combined with a structural model and the assumption of equilibrium behavior to

identify the effect. Interestingly our results suggest that the increase from mostly 2 bidders to mostly

3 bidders reduces the markup. By contrast Hong and Shum find going from 2 to 3 bidders actually

increases rents going to bidders in many of the auctions they study; Li and Zheng, on the other hand,

find negative effects of competition only when there are more than 3 bidders.

  The remainder of the paper proceeds as follows. Section II describes the setting and research

design. Section III explores the bidding process impact under both regular and enhanced competition.

  

Section IV explores the degree to which capture by vested interests reduced the impact of outsourcing.

Section V concludes.

II. SETTING, EXPERIMENTAL DESIGN, AND DATA

A. Setting

  

We examine Indonesia’s subsidized rice program, known as “Raskin” (Rice for the Poor). First

introduced in 1998, the program entitles 17.5 million low-income households to purchase 15 kg of rice

per month at a co-pay of Rp. 1,600 per kg (US$0.15), or about one-fifth of the market price. The

intended subsidy is substantial, about 4 percent of a beneficiary households’ monthly consumption.

It is Indonesia’s largest permanent, targeted social assistance program, with an annual budget of over heterogeneity in where they distribute it—at the village head’s office, at the homes of hamlet or neighborhood heads, or even directly to beneficiaries’ houses. Local governments are not only responsible for the time and effort required to distribute the rice, but they also assume the transportation

  8 costs, which in control areas cost an average of Rp. 244,161 (US$21) each month.

  In practice, Raskin faces a number of challenges, many of which occur in the last mile of service delivery. Rice may go missing at all stages in the distribution chain—from the central government to the sub-district distribution point to within hamlets. Evidence suggests, however, that many of the issues with missing rice crop up in the last mile of service delivery: While only 1 percent of Raskin distributors in the sample report receiving less than the full village quota in the last month from the government logistics agency, household purchases reveal that a substantial share of the quota never reaches households at all (Olken 2006 estimates that at least 18 percent of rice goes missing; World Bank 2012 estimates around 50 percent). Moreover, the rice that does arrive may be given to ineligible households rather than the eligible ones. On top of this, as shown in Appendix Table 1, households often have to pay a higher copay price (Rp. 660 per kg, or about a 40 percent markup) than the central

  9 government intends. All of these factors reduce the value of the intended transfer.

  It is important to note that these facts do not necessary imply malfeasance: local governments may be diverting rice to deserving, but ineligible households, or they may charge a higher copay for legitimate reasons, for example, to cover the transportation costs of distributing the rice. However, the distributors in our control group report transport costs that only account for about 12.4 percent of the price markup reported by households. Thus, it is likely that much of the higher price and missing they waited too long before picking it up. Anecdotally, people complain that Raskin rice is often crushed and mixed with small stones, which is one way corrupt local officials disguise the weight of sold rice, or that nefarious officials sell official Raskin rice to private traders and replace it with lower quality rice.

  B. Sample

  This project was carried out in 6 districts in Indonesia (2 each in the provinces of Lampung, South Sumatra, and Central Java). The districts are spread across Indonesia—specifically, on and off Java— in order to capture important heterogeneity in culture and institutions (Dearden and Ravallion, 1988).

  To further capture heterogeneity across institutions, we ensured that the sample consisted of about 40 percent urban and 60 percent rural locations. Within these districts, we had originally randomly sampled 600 locations. Prior to conducting the randomization, we dropped 28 localities that were

  12 deemed too unsafe to send survey teams. Thus, the final sample comprised 572 localities.

  C. Experimental Design

  Stratifying by geographic location and the previous experiments, we randomly assigned the 572 locations to one of three treatment assignments—bidding, bidding with enhanced competition, and information-only—or to a control group, as follows: the current distribution process and then the procurement process would be announced. At this meeting, citizens were told that anyone who wanted to—from both within and outside the locality—could bid for the right to distribute Raskin by submitting a bidding form within 10 days. The bidding form was a standard one that was provided to the local government, which included, but was not limited to, the price that the prospective bidder would charge citizens, the process (e.g. where the rice would be distributed, whether the households would have to pay upfront), and the bidder’s qualifications (e.g. access to credit, owning a truck). The central government insisted that households should receive their full allotment of rice, so the quantity of rice that the potential distributor would allow households to buy was not included on the forms. Bidders did not necessarily know the number of other bidders when they submitted and the bids remained sealed until the bidding meeting. Individuals were told that the winner would have the right to distribute Raskin for 6 months, with another meeting held at that time in which the committee would decide whether to continue with him, revert to the previous distributor, or set up a new bidding process.

  In addition, a small committee was formed during this organizational meeting to oversee the bidding process and to monitor its outcomes. The committee included members of the independent local monitoring committee (the Lembaga Pemberdayaan Masyarakat, Agency for Community Empowerment, “LPM”) charged with overseeing community development and improving the quality of local public services, neighborhood heads, informal community leaders, and Raskin beneficiaries. To avoid conflicts of interest, current distributors were excluded from being on this committee.

  13

  a public meeting. If more than five bids were submitted (which only happened in 7 locations), only the best five were to be presented at the meeting to ensure sufficient time for discussion. Although the facilitator took notes at the meeting, their participation was minimal and a committee representative led the meeting. During each presentation, the key proposal information was written on a large notepad to facilitate discussion. Bidders were allowed to improve upon their bids during the meeting in response

  14 to questions or in response to other bids.

  After the presentations, the committee members privately scored each proposal according to their criteria and summed the scores to determine the winner. Each bid was scored with a 1-10 qualitative score on each dimension, so that committees de facto had substantial leeway in how they assessed various bids. The committees always had an odd number of members (3 or 5) to ensure no ties. They also had the option of rejecting all of the bids and reverting to the status quo if they deemed that none were of high enough quality. At the end, the village head issued a letter establishing the winner as the official distributor for the next six months; this letter was also provided to relevant sub-district and district officials so that the winner could pay for and pick up the Raskin rice at the warehouse.

  The facilitators returned to the locality about six months later. At this time, the current distributor made a presentation about the Raskin distribution process as it operated at that time and the committee discussed their views on the process. They also decided whether to extend the new winner’s time as Raskin distributor (if there was one), to choose a new distributor either using the same bidding process or another method of their choosing, or to revert back to the old process. In this context, three bidders versus two bidders can potentially change the overall composition of bidders in important ways. When there are only two bidders and one of the bidders is the potentially inefficient incumbent government distributor, then the challenger needs only to beat the government distributor’s price. When there are three bidders, however, then the second best bidder may be another “efficient” (i.e. non-government) bidder, potentially putting greater pressure on lowering prices. This treatment thus served as a randomized increase in the number of bidders, though it did not necessarily require three bidders if indeed three bidders could not be found.

  

Information-Only: T he bidding process naturally provides greater transparency: one must provide

  information about the distribution process, so that potential bidders can decide whether to participate and , if so, can prepare realistic bids. But, the act of simply being forced to publicly itemize costs might lead distributors to lower markups if they could not provide adequate justification for their costs or if citizens notice a discrepancy between reported costs and price markups. Thus, any observed effects could be driven by greater transparency.

  To control for the information effects of the bidding treatment, we also randomly selected 96 locations for an information-only treatment, where a community facilitator coordinated with the village head to set up the organizational meeting. This meeting exactly mimicked the meetings held in the bidding villages, following the same procedures for setting up a committee comprised of LPM members, informal community leaders, neighborhood heads, and Raskin beneficiaries tasked with

  D. Randomization Design, Timing, and Data

  Appendix Table 2 shows the number of locations randomly assigned to each treatment. We stratified by 6 geographic strata (districts) and the previous experimental treatments.

  The timeline was as follows (Appendix Figure 2): in April-July 2013, after the baseline survey was completed for the entire sub-district, both treatments were conducted. During the following six months, facilitators maintained a call center to address any on-the-ground issues; only 17 calls were ever received. In January-February 2014, after the endline survey was completed in that sub-district, the facilitators returned to hold the follow-up meetings.

  E. Data Collection

  An established, independent survey organization (SurveyMeter) conducted the surveys. Two household surveys serve as our baseline, one conducted in October and November 2012 and one in April and May 2013. Each survey was conducted in a separate randomly-selected sub-unit (RW) within the locality. In total, across both survey waves, we randomly sampled between 15 and 19 households in each locality,

  17

  for a total of 10,277 households. We surveyed the households on their background and their Raskin experiences. At this time, we additionally interviewed the village head.

  In December 2013 and January 2014, just before the six-month follow-up meetings were held in the treatment locations, an endline survey took place in which we interviewed 6 randomly-selected households from each of the two baseline surveys (12 households per location), for a total of 6,864 households. As in the baseline surveys, we also surveyed the village head.

  

cheating statistically by looking for scores that are higher than would be predicted by chance. Hanna

and Wang (2015) show that this task is correlated with real-world corruption: they show that a high score is correlated with fraudulent absenteeism by government nurses in India.

  Finally, we have access to administrative data from the bidding forms filled out by prospective bidders and facilitators of the bidding process.

  F. Experimental Validity

Appendix Table 6A provides a check on the randomization of locations to the control, bidding and

information treatments. We provide the difference, conditional on strata, between bidding and pure

control (Column 5), information-only and pure control (Column 6), and bidding and information-only

(Column 7). Of the 45 differences that we estimate between the groups, only 5 (11 percent) are

significant at the 10 percent level, which is consistent with chance. The joint p-value across all

15 variables is 0.7, 0.50, and 0.20 in Columns 5-7, respectively. In Appendix Table 6B, we also

conduct a randomization check on enhanced competition versus the open bidding process. Again,

the two treatment groups appear balanced with none of the individual differences statistically significant at the 10 percent level and with a p-value for a joint significance test of 0.68.

  G. Descriptive Statistics on the Bidding Process

In Figure 1, we document the flow of the 191 bidding locations through the process. We also provide

the average Raskin price markup reported in both the baseline and endline household surveys at each

step

  Table 1 presents descriptive statistics on the bidding process. In Column 1, we present the overall mean, while in Columns 3 and 5, respectively, we present the means for locations randomly assigned either to the regular or enhanced competition bidding process. In Column 7, we present the p -value of the difference of means across the regular bidding process and enhanced competition.

  Citizens did bid for the distribution rights (Panel A). On average, we observed 2.43 bids placed, with 2.16 passing the initial screening process by the local committee and thus considered at the meeting. However, the process may have been dominated by the opinions of a few, namely

  18

  the elites (Panel B of Table 1). On average, about 22 individuals attended the bidding meetings (the average locality size is 1,299 households). Local leaders comprised a fair share of the participants, with about 9 of them attending, on average. About 8 of the meeting participants claimed to be Raskin beneficiaries. The facilitators reported that relatively few people spoke at the meetings, with no discussion from the crowd in 9 percent of the meetings and with less than 10 percent of attendees talking at 43 percent of them (Panel C). In only 3 percent of the meetings did they report that more than half of the crowd participated.

  The enhanced competition treatment led to more legitimate bids considered at the meeting, but did not change the probability of selecting a new distributor (Panel A). There were 2.74 bids in locations randomized to the enhanced competition treatment as opposed to 2.14 without the requirement, about a 30 percent increase; this difference is significant with a p-value of 0.01. One worry is that to fulfill the requirement, we would observe more “unrealistic” or “ghost” bids, but this was not the case: in the enhanced competition areas, we observe an increase in bids that pass the screen (2.44 relative to 1.88;

  • =
  • ( )

    where i represents a study location and s represents one of our geographic strata. The dependent variable

    in each column is a different characteristic of the distributor at endline (approximately six months

    after the intervention); this specification, thus, captures the net intent-to-treat effect of the treatment,

    including the fact that bidding may not always have occurred, that distributors may naturally change

    over time, and that the winning bidders may be blocked, resign, or be otherwise forced out. We include

    an indicator variable for whether there was either the bidding or information-only treatment

    ) ).

  Thus, (( ) and an indicator variable for just the bidding treating ( the coefficient captures how the bidding locations differ from those that received the information-

only (i.e. placebo) treatment and is the key coefficient of interest. We also report the p-value of the

difference of the bidding treatment against the pure control group (i.e. a test of + = 0) in the row

labeled “Bidding = Ctl”).

  Second, we separately estimate the effect of bidding with and without enhanced competition: =

  • ( ) _ + _ +

    In this regression, estimates the impact of the regular bidding procedure relative to the information

    placebo, and estimates the impact of the bidding procedure with the enhanced competition treatment

    (i.e. where a minimum of three bids were encouraged), relative to the information placebo. In this

    specification, we also report p-values of the difference between regular and enhanced bidding (i.e. a p-

  

to be in charge (Column 2), but their spouses/relatives and hamlet level-leaders were then more likely

to be in charge (Columns 3 and 4); thus, overall elite participation after the bidding process was not

greatly different than in the pure control group. Interestingly, this same pattern was occurring in the

information-only group as well, and while the effects are qualitatively bigger in the bidding group than

the information-only group, the differences are not statistically significant. This suggests that some of

the change in leadership may have been due to greater information.

  The more noticeable change was that there was a large increase in the probability that the

distributor was a trader by occupation in the bidding areas, relative to both the information-only and

pure control group (Column 5). Traders are likely to have skills and assets relevant to distributing

Raskin, though they are perhaps more likely to be a part of the “elites.” In short, while the bidding

treatment changed the identity of those distributing Raskin, it largely redistributed the role within the

existing local government elite. However, within the elite, it reallocated the job to people with the

relevant experience as a trader. Both bidding treatments produced broadly similar results on these

dimensions .

B. Impact on program outcomes

  

Did the bidding process change actual program outcomes and satisfaction? In Table 3, we focus on

outcomes from the household survey data. We estimate the same equations as in Table 2 using OLS,

but now cluster the standard errors to account for fact that the randomization was conducted by locality.

  In Table 3 Panel B, we separately identify the price effect for the regular bidding treatment and that with enhanced competition. Here, we find quite stark results: we only see price reductions in localities with enhanced competition. Specifically, households in enhanced competition localities pay

  Rp. 74 less than in the information placebo, about an 11 percent reduction in markup; this reduction is also statistically different from the pure control group. Households in localities with regular bidding pay a statistically insignificant Rp. 23 less than the information placebo, and only Rp. 5 – less than 1 percent – less than the pure controls. These results suggest that competition helps achieve price reductions – the opportunity to outsource itself is not enough. We return to this finding in more detail below.

  One worry is that to compensate for the lower price, more rice would go missing. This may particularly be the case because as the central government had mandated that distributors were supposed to provide the correct quantity of rice—and provide it only to eligible households—so this was not a category in the application form for the bid and therefore not a criterion on which bidders were evaluated, even though correct distribution is important in practice. Put another way, since all distributors were in theory supposed to distribute all the rice, it was not possible for bidders to compete on this dimension. In any case, the overall quantity of rice bought did not change (Column 3).

  The key concern articulated by the HSV theory is that, as a result of outsourcing, private distributors may shirk and reduce non-contractible dimensions of quality. In our case, a key dimension is rice quality. Distributors can increase quality by refusing to accept low quality deliveries from the warehouse or by stopping a practice of selling high quality rice on the market and substituting lower

  

process across the treatments (Column 8). Overall satisfaction actually fell in the information treatment

as citizens learned more about how the process should really look, with no additional difference for just

the bidding process.

  Overall, we observe a decrease in the price markup, but only in areas with enhanced competition. We find no evidence of a decline in quality.

C. Impact on Distribution Costs

  

As HSV point out, the theory tells us that even when contracting out leads to efficiency gains, without

sufficient competition, these efficiency gains may be captured by vendors rather than enjoyed by

the public at large. To investigate these issues further, at the endline, we interviewed the distributor

(whomever it was) and asked about their distribution costs. Note two aspects of the cost measures. First,

they are self-reported; given the informal nature of the economy, one cannot track them through credit

card or bank transactions. Nevertheless, they may shed light on how the distributers functioned. Second,

the reported costs often increase in the information treatment relative to the pure control, likely because

it

forces distributors to better compute their actual costs (and because they may be constrained to make

sure the costs add up to the total markup) or because the greater scrutiny forces them to report their true

costs. Given this, it is important to compare bidding and information to information-only, rather than

to pure control, to hold this transparency effect constant.

  Table 4 shows that, indeed, we observe a decrease in transportation costs in the bidding

treatment overall, relative to just pure information (Column 1). These reductions seem roughly similar

  We find that most of the difference in prices from increased competition comes from differences in behavior of distributors ex-post, rather than coming from lower-priced bids being submitted ex-ante. While standard auction theory suggests that bidders in a sealed-price first-price auction should bid more aggressively if they expect more bidders (e.g. Milgrom and Weber 1982), we find no evidence of this in our context (Appendix Table 22). We do, however, find that bid committees choose differently among bidders when they have more choice. Specifically, with more choice, bidding committees appear to place more weight on factors like living in the locality or experience as a trader relative to price (see the Appendix for more details.) And, although the estimates are somewhat noisy, the data also suggest that the actual price paid is closer to the promised price in the enhanced competition treatment than in the regular bidding treatment (Appendix Table 24). This could be because the winners are more reliable, or perhaps the enhanced competition treatment allowed the village to discover more potential suppliers, leading to greater pressure on the chosen supplier to stick to their promise .

  On net, the experimental results presented in this section point to the importance of sufficient ex

  • ante competition for ensuring the success of outsourcing: outsourcing without sufficient competition can yield efficiency improvements, but sufficient competition is needed to ensure that the gains are passed on to the government.

IV. THE ROLE OF EXISTING RENTS: THE RACE BETWEEN VESTED

  baseline (Columns 1 and 2) and the Raskin price at endline in our control areas (Column 3-5) various corruption metrics, controlling for local characteristics that proxy for actual transportation costs

  23

  (e.g. distance to the sub-district, log population, and number of hamlets). Higher Raskin prices are not only strongly correlated with citizen perceptions of corruption, but are also positively correlated with the distributors scoring higher than median on the experimental dice-based dishonesty task. This suggests that the baseline price may indeed capture not just operating costs, but also the amount of rents in the system.

  Table 6 then examines whether failure of the process ex-ante or ex-post appears correlated with baseline prices and corruption of the incumbent Raskin distributor, as proxied by their score on the dice task . To the extent that failure of the process (e.g. no bidding meeting held) is positively correlated with baseline prices, this suggest blocking by elites seeking to protect vested interests; to the extent that it is negatively correlated with baseline prices, this suggests the system is responding appropriately, with villagers not bothering to outsource when the system is working well.

  Table 6 begins by investigating whether privatization occurs or not– i.e. of the 191 locations randomized to bidding, in which types of areas did bidding actually occur? We regress a dummy variable that equals 1 if there was no bidding meeting or no bids at the meeting on local characteristics. E ach cell in Column 1 comes from a separate regression; Columns 2 – 4 report the results from a single

  24

  regression in each column. Higher prices substantially predict the occurrence of a meeting with at least one bidder: a one standard deviation increase in baseline markup (Rp. 390) would increase the log-odds by 1.31 i.e. increasing the odds of having the meeting with at least one bidder by over

  The remaining columns of Table 6 examine other points in the bidding process where vested interests might exercise some control. Columns 5-8 investigate the selection stage by examining the probability that the incumbent distributor was chosen as the winner, conditional on the bidding process occurring (i.e. conditional on it not being blocked in the first stage; results defined for all 191 treatment locations are available in Appendix Table 14). We find similar countervailing forces at play at the selection stage as well. The incumbent is less likely to be chosen when baseline prices were high, though the results are about a third of the magnitude as in the previous table. The incumbent is also

  25

  more likely to be chosen when baseline household satisfaction is high. However, this is also offset by the fact that dishonest incumbents (as measured by dice score) are more likely to win.

  The final set of results in Table 6 examines whether the incumbent distributor is still distributing six months later, conditional on him not having won the bidding. This variable captures ex-post capture. Here, very little predicts action at this stage, suggesting that on average at least, most of the tussle over rents happens before and during the bidding, not after.

  Tables 6 and 7 show countervailing forces at play during the bidding process: high rents leads to more entry, but corrupt officials are more likely to block the process, either by preventing the bidding from occurring or by selecting the incumbent from among the bids provided. The experimental estimates from Section 3 suggest that the additional entry induced by the high baseline rents should lead to more substantial price reductions. For example, the estimates from Table 7 imply that moving from th th the 10 to the 90 percentile in baseline price markup (i.e. from Rp. 129 to Rp. 1359) would lead to an ad ditional 1.37 bidders, which is more than twice as large as the effect of the enhanced competition th

  Rp. 90 overall, and RP. 119 for the enhanced competition treatment – at the 90 percentile of the distribution. This suggests that on net the treatment was most effective in more problematic areas, despite the fact that vested interests may have had more rents to protect. th th

  The difference we estimate using the quantile treatment effects between the 10 and 90 percentile – about Rp. 80/kg overall, and about Rp. 110/kg for the enhanced competition treatment – is almost precisely what one would predict based on the additional entry induced by the difference in baseline prices. This suggests that the additional entry induced by the high prices may have been critical to the heterogeneous treatment effects we observe.

  All told, the evidence presented suggests offsetting effects. On the one hand, we do see evidence that in areas with high rents, corrupt elites appear to have been working to protect their rents either ex -ante by preventing people from bidding or during the process by maneuvering to have themselves selected as the winner of the bidding. On the other hand, we see results consistent with upward sloping supply curves: the bidding was more likely to be held and there were more bidders in areas where there were more rents to begin with. And, as shown in Section 3, adding additional bidders seems to reduce prices substantially. On net, the entry effect and demand effect dominates, and the opportunity to outsource had the largest effects in areas where baseline rents were highest.

V. CONCLUSION

  In this paper, we examine whether allowing local governments to outsource the delivery to the private sector improves the distribution. Focusing on a subsidized food distribution program, we show that outsource in the areas with high baseline rents, the power of vested interests to block may have muted the effect of outsourcing to some extent. More generally, this suggests while there can be gains from outsourcing, it is important to think about how we design these processes in contexts with strong vested interests.

WORKS CITED

  Acemoglu, Daron, and James A. Robinson. “Political Losers as a Barrier to Economic Development.” American Economic Review 90, no. 2 (2000): 126-130. Alatas, Vivi, Abhijit Banerjee, Rema Hanna, Benjamin A. Olken, and Julia Tobias. “Targeting the Poor:

  Evidence from a Field Experiment in Indonesia.” American Economic Review 102, no. 4 (2012): 1206-40. Alatas, Vivi, Abhijit Banerjee, Rema Hanna, Benjamin A. Olken, Ririn Purnamasari, and Matthew Wai- poi. “Self-Targeting: Evidence From A Field Experiment In Indonesia.” Journal of Political

  Economy , forthcoming.

  Bajari, Patrick, Robert McMillan, and Steven Tadelis. “Auctions Versus Negotiations in Procurement: An Empirical Analysis.” Journal of Law, Economics, and Organization 25, no. 2 (2009): 372-399. Bajari, Patrick, Stephanie Houghton, and Steven Tadelis, “Bidding for Incomplete Contracts:

  An Empirical Analysis of Adaptation Costs.” The American Economic Review 104, no. 4 (2014): 1288-1319. Bajari, Patrick, and Steven Tadelis. “Incentives Versus Transaction Costs: A Theory of Procurement Contracts.” RAND Journal of Economics 32, no. 3 (2001): 387-407. Bandiera, Oriana, Andrea Prat, and Tommaso Valletti, “Active and Passive Waste in Government

  Spending: Evidence from a Policy Experiment,” American Economic Review 99 (2009), 1278-1308. Banerjee, Abhijit, Rema Hanna, Jordan C. Kyle, Benjamin A. Olken, and Sudarno Sumarto. “Tangible Decarolis, Francesco. “Awarding Price, Contract Performance, and Bids Screening: Evidence from Procurement Auctions.” American Economic Journal: Applied Economics 6, no. 1 (2014): 108-132.

  Dearden, Lorraine, and Martin Ravallion. “Social Security in a ‘Moral Economy’: An Empirical Analysis for Java.” Review of Economics and Statistics 70 (1988): 36-44. Fischbacher, Urs, and Franziska Föllmi‐Heusi. “Lies in Disguise: An Experimental Study on Cheating,” Journal of the European Economic Association 11 (2013): 525-547. Gil, Ricard, and Jean-Michel Oudot. “Competitive Bidding, Renegotiation and Relational Contracting: Evidence from French Defense Procurement.” Working Paper (2009). Government of Indonesia. 2012. “Nota Keuangan dan Rancangan Anggaran Pendapatan dan Belanja

  Negara Perubahan tahun anggaran 2012 [Financial Note and Revised Budget 2012].” http://www.perpustakaan.depkeu.go.id/ FOLDERDOKUMEN/Th.%202012%20perubahan.pdf Gustafsson, Bjorn A and Deng Quheng. “Di Bao Receipt and Its Importance for Combating Poverty in Urban China.” Poverty & Public Policy 3 (1), Article 10, (2011). Hanna, Rema, and Shing-Yi Wang. “Dishonesty and Selection into Public Service: Evidence from India.” NBER Working Paper No. 19649 (2015). Hart, Oliver, Andrei Shleifer, and Robert W. Vishny. “The Proper Scope of Government: Theory and an Application to Prisons.” Quarterly Journal of Economics 112 (1997): 1127-1161. Hirsch, Werner Z. “Factors Important in Local Governments' Privatization Decision.” Urban Affairs

  Quarterly 31 (1995): 226-43. Sukhtankar, Sandip. “The Impact of Corruption on Consumer Markets: Evidence from the Allocation of 2G Wireless Spectrum in India.” Journal of Law and Economics 58 (2015). Spulber, Daniel F. “Auctions and Contract Enforcement.” Journal of Law, Economics, and Organization (1990): 325-344. Tran, Anh, “Which Regulations Reduce Corruption? Evidence from the Internal Records of a Bribe- paying Firm”, mimeo, Indiana University (2011). World Bank. “Raskin Subsidized Rice Delivery: Social Assistance Program and Public Expenditure Review.” Jakarta, Indonesia: Memo (2012).

  APPENDIX

A. Selecting Among Bidders

  Selecting among bidders is complex. To the extent that there are non-contractible dimensions of quality, those running procurement procedures may wish to use characteristics that predict quality, such as a bidder’s reputation or other potential performance indicators. One may also be concerned that bidders may attempt to renegotiate ex-post if the winner cannot be forced to honor his original proposal; indeed, such renegotiations may be optimal in order to share risk if there is some information about the job that is only revealed ex-post. But, if renegotiations are allowed, a firm might adopt the strategy of bidding low to win and ask for better terms later (or abandon the job), ultimately leading to higher costs (Chang,

1 Salmon and Saral, 2013; Decarolis 2014). Others (e.g. Gil and Oudot (2009)) have recommended

  2 abandoning bidding altogether and negotiating with firms. T here may also be incompetence or laziness among those running the bidding, or even outright corruption (see, for example, Bandeira, Prat, and Valletti, 2009; Tran 2011; Sukhtankar 2015), resulting in substantial incumbency advantage.

  We start by comparing winning (Column 1) and losing bids (Column 3) in Appendix Table 9. In Column 5, we present the p-value of the difference between the two; note that this is clustered by location. Winning bids differ from losing ones on numerous dimensions. On average, the winners propose a lower markup (Rp. 472/kg) than the losers (Rp. 567 /kg); this 17 percent difference is significant at the 1 percent level. The average winning bid proposed a 28 percent lower markup than the baseline markup of Rp. 654/kg. These averages mask considerable heterogeneity. Appendix Figure for the village is recorded in the government bank account. On average, distributors need to remit a copayment of Rp. 9 million (US$960) each month in order to collect the village’s rice quota, a sum that could be difficult to amass without collecting payments in advance.

  Appendix Table 9 includes all bids, regardless of whether or not there was more than one bid. To more formally analyze how bidding committees selected among bids, we restrict our sample to areas with multiple bids and estimate a conditional logit discrete choice model in Appendix Table 10.

  In Column 1, we explore each bid characteristic one by one on the probability of winning (i.e., each cell reports the result from a separate univariate regression). In Column 2, we include all bid proposal characteristics jointly, and in Column 3, we also add the individual characteristics of the bidder to the

  3 specification in Column 2.

  As shown in Appendix Table 10, the proposed price is a significant predictor of winning, even conditional on other proposal (Column 2) or individual characteristics (Column 3), further implying that price enters strongly into the decision rule. In fact, the lowest bidder wins 82 percent of the time.

  We find no evidence of capture in the bidding process. While being the distributor at the time of bidding is also advantageous (Column 1), the effect becomes smaller in magnitude and insignificant when controlling for proposal characteristics, suggesting that this advantage is driven by being able to

  4 propose a more attractive bid, rather than an incumbency advantage per se.

  On the other hand, there is evidence that, even conditional on price, the committees select bidders with skills that may make them more effective distributors. Specifically, bids that come from traders have an advantage, even conditional on other bid characteristics. The committee also appeared attempting to solve the quality or ex-post default problem by choosing those with low unobserved costs or high skills, with the idea that conditional on price they may deliver higher quality and not default on their commitments.

  Appendix Table 11 examines the degree to which the enhanced competition treatment changes the bidding committee’s decision rule. Specifically, we re-estimate Appendix Table 10, interacting each variable with the enhanced competition treatment. In the enhanced competition treatment, we find that citizens prefer the candidates who promise that they do not have to pay before receipt, and those with trading experience and transportation access, and those who live in the locality, but do not find an observable difference in choosing on price markup. Thus, the increase in competition seems to allow committees to exercise preferences over aspects of the bid other than pure price. With more choice, bidding committees appear to choose a candidate who is more reliable on other dimensions for the same promised price, and that these candidates actually deliver on what they promise, rather than ex-post reneging and channeling profits to amorphous ‘compensation to others’ (Table 4).

  We also test whether the additional competition changes the bids themselves (Appendix Table 22). While standard auction theory suggests that bidders in a sealed-price, first-price auction should bid more aggressively if they expect more bidders (e.g. Milgrom and Weber 1982), we find no evidence of this in our context. Specifically, in Columns 1 and 2, we find no evidence that the incumbent reduced their bid in response to the minimum number of bidders treatment, and in fact the point estimates are positive, albeit noisy. Columns 3 and 4 show that the average bid also did not change. Even more surprisingly, Columns 5 and 6 show that the minimum price bid did not change, though again estimates

  Table 1: The Bidding Process (Conditional on Bidding Meeting Occurring) Overall Regular Bids Enhanced Comp. P-Value Regular =

  Enhanced

Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.

(1) (2) (3) (4) (5) (6) (7)

  Panel A: Bids Submitted Number of Bids

  2.43

  1.66

  2.14

  1.68

  2.74 1.59 0.01** Number of Bids, After Initial Screening

  2.16

  1.50

  1.88

  1.47

  2.44 1.48 0.01** Old Distributor Wins

  0.52

  0.50

  0.49

  0.50

  0.55

  0.50

  0.49 Panel B: Meeting Attendance Attendees

  21.69

  9.13

  21.09

  10.31

  22.31

  7.75

  0.36 Raskin Beneficiaries

  8.28

  8.32

  8.78

  8.95

  7.77

  7.63

  0.41 Local Officials

  9.42

  5.83

  8.80

  5.95

  10.06

  5.66

  0.14 Panel C: Meeting Participation No Discussion at Meeting

  0.09

  0.29

  0.03

  0.18

  0.15 0.36 0.01*** <10% of People Talk

  0.43

  0.50

  0.46

  0.50

  0.41

  0.49

  0.51 10-50% of People Talk

  0.45

  0.50

  0.50

  0.50

  0.40

  0.49

  0.16

  30 >50% of People Talk

  0.03

  0.16

  0.01

  0.11

  0.04

  0.21

  0.18 Notes: This table provides summary statistics on the number of bids submitted, as well as the attendance and participation during the bidding meeting. All data come from the forms that the facilitators used to document the bidding process. We first present the sample statistics for the 184 localities where a bidding meeting was held and then we disaggregate the data by whether the locality was randomly assigned to the minimum bid requirement (91 localities) or it was left open (93 localities). *** p<0.01 ** p<0.05 * p<0.1.

  

Table 2: Who Distributes Raskin Six Months After Intervention?

In charge of any Distributor/spouse is Distributor/spouse responsibilities Distributor/spouse related to a local is/was or is related to before May 2013 is/was local official official hamlet official Is a trader Lives in locality (1) (2) (3) (4) (5) (6)

  

Panel A: Combined Effect of Outsourcing

Info or Bidding -0.035 -0.059 0.043 0.001 0.008 -0.027 (0.048) (0.050) (0.044) (0.054) (0.027) (0.042) Bidding -0.165*** -0.076 0.046 -0.002 0.071** -0.019 (0.051) (0.053) (0.047) (0.058) (0.028) (0.045) P-Value Bidding = Ctl 0.000 0.001 0.013 0.970 0.000 0.175 Control Mean 0.803 0.347 0.160 0.350 0.034 0.816

  

Panel B: Estimating Additional Effect of Competition

Info or Bidding -0.035 -0.059 0.043 0.001 0.008 -0.027 (0.048) (0.050) (0.045) (0.054) (0.027) (0.042) Regular Bids -0.135** -0.110* 0.053 0.036 0.068** -0.046

  31 (0.059) (0.062) (0.055) (0.067) (0.033) (0.052) Enhanced -0.197*** -0.042 0.038 -0.041 0.074** 0.009

  Competition (0.059) (0.062) (0.055) (0.067) (0.033) (0.052) P-Value Regular = Enh 0.297 0.272 0.783 0.248 0.839 0.288 Regular = Ctl 0.001 0.001 0.033 0.499 0.005 0.085 Enh = Ctl 0.000 0.046 0.073 0.460 0.002 0.671 Control Mean 0.803 0.347 0.160 0.350 0.034 0.816 Note: In this table, we explore the characteristics of bidders across the experimental groups, six months after the intervention. We regress each

characteristic on indicator variables for the bidding and information treatments and strata fixed effects. In Panel B, we disaggregate the bidding effect by

whether the locality was randomized into the minimum number of bids requirement. All regressions are estimated by OLS. Each column in each panel

  Table 3: Raskin Distribution Process Distance to Time to Amount Satisfied with purchase point purchase point Paid for rice in Satisfied with

  

Bought Raskin Price markup purchased rice quality (meters) (minutes) advance Raskin program

(1) (2) (3) (4) (5) (6) (7) (8)

Panel A: Combined Effect of Outsourcing

  Info or Bidding -0.009 18.770 0.151 0.006 -1.080 0.441 0.013 -0.020*

(0.02) (24.07) (0.23) (0.01) (14.37) (0.27) (0.02) (0.01)

Bidding 0.021 -49.023** -0.002 0.019* 7.754 -0.501* -0.009 0.006

(0.03) (24.91) (0.24) (0.01) (15.15) (0.28) (0.03) (0.01)

P-Value Bidding = Ctl

  0.55

  0.09

  0.39

  0.00

  0.55

  0.75

  0.82

  0.06 Observations 6,860 5,886 6,858 6,533 6,194 6,247 6,394 6,782 Control Mean 0.76 652.39

  5.76

0.51 190.96

  5.94

  0.43

  0.59 Panel B: Estimating Additional Effect of Competition Info or Bidding -0.009 18.842 0.151 0.006 -0.999 0.441 0.013 -0.020*

(0.02) (24.06) (0.23) (0.01) (14.38) (0.27) (0.02) (0.01)

Regular Bids 0.014 -23.645 -0.016 0.016 25.315 -0.430 0.011 0.001

  32

(0.03) (29.94) (0.28) (0.01) (18.11) (0.33) (0.03) (0.01)

Enhanced 0.027 -73.551*** 0.012 0.022 -9.368 -0.569* -0.028 0.011

  Competition (0.03) (26.51) (0.27) (0.01) (16.62) (0.31) (0.03) (0.01) P-Value Regular = Enh

  0.66

  0.06

  0.92

  0.66

  0.04

  0.63

  0.17

  0.38 Regular = Ctl

  0.84

  0.84

  0.55

  0.05

  0.10

  0.96

  0.29

  0.06 Enh = Ctl

  0.46

  0.01

  0.44

  0.01

  0.44

  0.57

  0.52

  0.29 Observations 6,860 5,886 6,858 6,533 6,194 6,247 6,394 6,782 Control Mean 0.76 652.39

  5.76

0.51 190.96

  5.94

  0.43

  0.59 Note: This table explores the effect of the treatments on the actual program functioning. All data come from the household endline survey that we

  

Table 4: Endline Costs to Current Distributor

Compensation to Transportation costs others Other costs Total costs

  (1) (2) (3) (4)

Panel A: Combined Effect of Outsourcing

  Info or Bidding 88,038* 121,875 40,716** 318,287 (52,052) (174,163) (18,745) (211,403) Bidding -101,616* -94,256 -30,531 -317,960 (54,924) (179,950) (19,678) (219,985) P-Value Bidding = Ctl 0.695 0.836 0.445 0.998 Observations 574 574 574 574 Control Mean 244,161 961,974 84,166 1,315,030

  

Panel B: Estimating Additional Effect of Competition

Info or Bidding 87,943* 123,544 40,726** 320,069 (52,102) (174,286) (18,740) (211,657) Regular Bids -124,126** 154,902 -13,578 -51,869 (59,089) (222,663) (22,454) (263,152) Enhanced -79,174 -349,228* -47,098** -590,262*** Competition (62,837) (180,668) (21,563) (223,414) P-Value Regular = Enh 0.396 0.009 0.093 0.013 Regular = Ctl 0.373 0.137 0.113 0.189 Enh = Ctl 0.850 0.095 0.693 0.084 Observations 574 574 574 574 Control Mean 244,161 961,974 84,166 1,315,030 Note: This table explores the effect of the treatments on the program costs (in Rp). All data come from the

  

Table 5: Corruption on Locality Price Markup

Baseline Raskin Markup Endline Raskin Markup (1) (2) (3) (4) (5)

  Raskin Distributor's Dice Score 61.029* 81.821** Above Median (31.315) (40.940) Perception of Local Head's 598.602*** 632.064*** Corruption

  (123.065) (179.431) Perception of Raskin 647.773*** Distributor's Corruption

  (210.777) Observations 454 455 282 283 273

Note: In this table, we explore the relationship between corruption and Raskin price markup. In each column,

we regress average price markup reported by households in a locality on a measure of corruption, controlling

for baseline locality characteristics (number of hamlets, log number of households, distance to subdistrict). Our

measures of corruption, by column, are: 1. Dummy for baseline distributor's dice score above median; 2. Average perception of locality head's corruption at baseline; 3. Dummy for endline distributor's dice score

above median; 4. Average perception of locality head's corruption at endline; 5. Average perception of Raskin

distributor's corruption at endline. Measures of corruption consist of 4 categories, scaled 0-1, where 0 equals "no possibility of corruption" and 1 equals "very high possibility of corruption." Columns 1 and 2 examine baseline Raskin markup and include all localities; Columns 3, 4, and 5 examine endline Raskin price and include control localities only. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

  

Table 6: When Did Original Distributor Win and Continue Distributing?

Where Was No Bidding Meeting Held, or Meeting Had No Where Did Original Distributor Win? (Conditional on Where is Original Distributor Still Distributing? Bids? Bidding Held with 1+ Bid) (Conditional on Not Winning) Joint Joint Joint (Household Joint (Form (Household Joint (Form (Household Joint (Form

  1-by-1 Only) Only) Joint (All) 1-by-1 Only) Only) Joint (All) 1-by-1 Only) Only) Joint (All) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Panel A: Reported by Households in Baseline

Avg Price Markup (Rp/kg) -0.00299*** -0.00291** -0.00337*** -0.00115** -0.00111* -0.00141 0.00017 -0.00011 0.00062 (0.00089) (0.00115) (0.00122) (0.00056) (0.00065) (0.00099) (0.00079) (0.00090) (0.00144) HH Bought Raskin in Last 2 0.369 0.517 0.285 1.598** 0.995 0.698 -0.206 0.392 1.493

  Months (0.89) (1.24) (1.46) (0.65) (0.80) (1.05) (0.87) (1.13) (1.79) Avg Amount of Raskin 0.0240 -0.0429 -0.0615 0.0606 -0.0031 0.0223 -0.0814 -0.0694 -0.121 Purchased (kg) (0.068) (0.091) (0.120) (0.071) (0.061) (0.064) (0.071) (0.088) (0.15) Avg Satisfaction with Program 3.887* 1.408 0.766 4.082** 3.987** 5.981** 0.524 -0.042 0.405 Quality (0-1 scale) (2.24) (2.99) (3.37) (1.77) (1.9402) (2.35) (2.27) (2.7635) (4.23) Avg Distance to Purchase Point 0.00078 -0.00070 -0.00030 -0.00181* -0.00234* -0.00300** 0.00163 0.00182 0.00104 (meters) (0.0013) (0.0015) (0.0016) (0.0010) (0.0012) (0.0014) (0.0013) (0.0014) (0.0018) HH purchased Raskin in 1.47*** 1.10** 1.20** 0.80** 0.24 -0.11 -1.10* -1.00 -1.28 advance (0.49) (0.53) (0.61) (0.38) (0.42) (0.51) (0.63) (0.77) (1.15)

Panel B: From Facilitation Forms

Raw dice score above median 0.81* 0.95** 1.01** 0.98*** 0.98** 0.87** -0.22 -0.42 -0.66

  (0.46) (0.48) (0.49) (0.37) (0.39) (0.43) (0.60) (0.62) (0.66)

  35 Old Distributor Provides Credit -0.69 -0.59

0.36 -0.26 -0.11

  0.24

  0.34 0.22 -0.44 if Recipient Cannot Afford (0.65) (0.73) (0.80) (0.39) (0.56) (0.66) (0.56) (0.83) (0.96) Costs of Rental Vehicle and/or -0.0062 -0.0065 -0.0120* 0.0011 -0.0023 -0.0000 -0.0025 0.0033 0.0047 Fuel to Old Distributor

  (0.0087) (0.0081) (0.0070) (0.0036) (0.0046) (0.0048) (0.0073) (0.0071) (0.0076) Non-Transportation Costs to -0.0003 0.0004 0.0025 -0.0013 -0.0008 -0.0000 -0.0021 0.0003 -0.0004 Old Distributor (0.0017) (0.0014) (0.0016) (0.0012) (0.0014) (0.0013) (0.0019) (0.0027) (0.0033) Joint P-Value 0.004 0.294 0.006 0.020 0.135 0.041 0.546 0.962 0.540 Observations 187 149 147 162 123 122

  76

  55

  54 Mean

  0.13

  0.17

  0.17

  0.53

  0.55

  0.56

  0.32

  0.33

  0.31 Note: In this table, we explore what characteristics predict that the locality with actually have bidders present at meeting, what characteristics predict that existing distributor will win (or that the committee will immediately throw out all the bids and return to the existing process), and what characteristics predict the continuation of the existing distributor's distribution. We regress a dummy for the existing distributor as the outcome of the bidding process on baseline

characteristics from the household survey (Panel A) and from the baseline information forms on process (Panel B). All regression are estimated as a logit. The top 1% of transportation and other costs are dropped; costs are reported in Rp

10,000. *** p<0.01 ** p<0.05 * p<0.1

  

Table 7: Baseline Markup and Number of Bids

Number of Bids (1) (2)

  

Baseline price markup 0.000895** 0.00111**

(0.000367) (0.000436)

Enhanced Competition 0.624** 0.615**

(0.243) (0.244)

Number of Hamlets -0.000764

(0.000584)

Log number of HH -0.0447

(0.204)

Distance to subdistrict -0.0249**

(0.0119) Observations 182 182

Note: In this table, we explore the relationship between number

of bids submitted and assignment to enhanced competition in

treatment localities. In Column 1 we control for baseline price

markup, in Column 2 we add baseline locality characteristics

(number of hamlets, log number of households, distance to subdistrict) as controls. *** p<0.01 ** p<0.05 * p<0.1

  

Table 8: Quantile Treatment Effects on Price Markup

Quantile 0.1 Quantile 0.25 Quantile 0.5 Quantile 0.75 Quantile 0.9 (1) (2) (3) (4) (5)

  

Panel A: Combined Effect of Outsourcing

Info or Bidding 0.000 7.705 -0.000 53.333 39.394 (15.52) (15.78) (12.07) (33.70) (41.26) Bidding -0.000 -7.705 0.000 -53.333 -90.505** (16.71) (15.78) (12.57) (34.06) (35.33) P-Value Bidding = Ctl

  1.00

  1.00

  1.00

  1.00

  0.13 Observations 5,886 5,886 5,886 5,886 5,886 Control Mean 150 400 600 900 1,114.29

Panel B: Estimating Additional Effect of Competition

  Info or Bidding 0.000 7.705 0.000 52.954* 45.722 (15.62) (15.85) (11.73) (30.73) (42.89) Regular Bids -7.843 -7.705 0.000 -47.709 -81.355* (17.56) (18.09) (15.72) (34.07) (42.48)

Enhanced -0.000 -7.705 -0.000 -56.751* -118.645***

Competition (18.77) (19.63) (13.40) (30.75) (41.28) P-Value Regular = Enh

  0.65

  1.00

  1.00

  0.59

  0.27 Regular = Ctl

  0.54

  1.00

  1.00

  0.69

  0.25 Enh = Ctl

  1.00

  1.00

  1.00

  0.76

  0.05 Observations 5,886 5,886 5,886 5,886 5,886 Control Mean 150 400 600 900 1,114.29

Note: This table explores the effect of the treatments on the actual program functioning. All data come

from the household endline survey that we conducted about six months after the intervention. We

1 Bid

  Endline: 599 Distributes At Least Once

  IDDING M EETING

  IDDING B

  P OST -B

  N: 5 Baseline: 686 Endline: 640

  Endline: 1,195 Blocked by District

  Blocked by Sub- District N: 2 Baseline: 1,164

  N: 4 Baseline: 1,091 Endline: 1,140

  Endline: 650 Blocked by Locality

  Blocked by Local Forum N: 1 Baseline: 538

  N: 11 Baseline: 800 Endline: 843

  Endline: 743 Blocked From Distributing

  Never Distributes N: 22 Baseline: 707

  N: 57 Baseline: 773 Endline: 738

  38 Figure 1: Flow of Localities through Bidding Process Randomized To Bidding

  N: 191 Baseline: 654 Endline: 637

  N: 79 Baseline: 754 Endline: 739

  N: 17 Baseline: 650 Winner is New

  Endline: 675 5+ Bids

  4 Bids N: 14 Baseline: 766

  Endline: 714

  3 Bids N: 58 Baseline: 706

  Endline: 642

  2 Bids N: 26 Baseline: 683

  N: 50 Baseline: 671 Endline: 620

  Endline: 386

  No Bids N: 20 Baseline: 370

  N: 6 Baseline: 624 Endline: 588

  Endline: 638 Reject Treatment

  Held Bidding Meetings N: 185 Baseline: 655

  Winner is Old N: 86 Baseline: 638

  Figure 2: Endline Markup

  1 .8 .6 .4 .2

  500 1000 1500 2000 Endline mean markup in village Enh. Comp. Reg. Bid.

  Contro/Info

  

Appendix Table 1: Baseline Summary Statistics

N Mean Std. Dev.

  (1) (2) (3)

  Appendix Table 3: Who Distributes Raskin Six Months After Intervention, Continued? Owns a truck and/or Avg digit span above Raw dice score above Has personal savings median median a boat

  Years of education account for business (1) (2) (3) (4) (5) Panel A: Combined Effect of Outsourcing

  Info or Bidding -0.039 -0.009 -0.021 0.233 0.036 (0.026) (0.059) (0.059) (0.325) (0.055) Bidding 0.022 0.002 0.023 -0.453 0.105* (0.028) (0.062) (0.062) (0.345) (0.058) P-Value Bidding = Ctl 0.432 0.875 0.963 0.397 0.001 Control Mean 0.071 0.449 0.519 12.139 0.313

  Panel B: Estimating Additional Effect of Competition Info or Bidding -0.039 -0.009 -0.021 0.233 0.036 (0.026) (0.059) (0.059) (0.326) (0.055) Regular Bids 0.027 0.020 0.048 -0.500 0.128* (0.032) (0.072) (0.072) (0.399) (0.068) Enhanced 0.017 -0.017 -0.002 -0.406 0.081

  40 Competition (0.032) (0.072) (0.072) (0.400) (0.068) P-Value Regular = Enh 0.747 0.609 0.492 0.816 0.485 Regular = Ctl 0.671 0.850 0.648 0.417 0.003 Enh = Ctl 0.414 0.662 0.700 0.598 0.036 Control Mean 0.071 0.449 0.519 12.139 0.313 Note: In this table, we explore the characteristics of bidders across the experimental groups, six months after the intervention. We regress each characteristic on indicator variables for the bidding and information treatments and strata fixed effects. In Panel

  B, we disaggregate the bidding effect by whether the locality was randomized into the minimum number of bids requirement. All regressions are estimated by OLS. Each column in each panel has 587 observations. *** p<0.01, ** p<0.05, * p<0.1.

  

Appendix Table 4A: Who Distributes Raskin Six Months After Intervention, by Women Quota Subtreatment

Panel A: Distributor Identity and Connections

In charge of any Distributor/spouse is Distributor/spouse responsibilities Distributor/spouse related to a local is/was or is related to Lives in before May 2013 is/was local official official hamlet official Is a trader locality Info or Bidding (1) (2) (3) (4) (5) (6)

  • 0.040 -0.132** 0.110* 0.050 -0.018 -0.033 No Women Quota (0.064) (0.066) (0.059) (0.072) (0.035) (0.056)
  • 0.030 0.015 -0.025 -0.049 0.035 -0.020 Women Quota Bidding

  (0.064) (0.066) (0.059) (0.072) (0.035) (0.056)

  • 0.157** 0.044 -0.027 -0.021 0.132*** -0.025 No Women Quota (0.072) (0.075) (0.067) (0.082) (0.040) (0.064)
  • 0.173** -0.197*** 0.119* 0.016 0.010 -0.013 Women Quota P-Value

  (0.073) (0.076) (0.067) (0.082) (0.040) (0.064) Info or Bidding: 0.904 0.092 0.081 0.297 0.256 0.862 No Quota = Quota

  Bidding: 0.877 0.024 0.126 0.749 0.032 0.892 No Quota = Quota

Control Mean 0.803 0.347 0.160 0.350 0.034 0.816

Panel B: Other Distributor Characteristics

Has personal

  Owns a truck Avg Digit Span Raw dice score savings account and/or a boat above median above median Years of education for business Info or Bidding (1) (2) (3) (4) (5)

  • 0.067* -0.005 -0.003 0.023 0.026 No Women Quota (0.035) (0.078) (0.078) (0.431) (0.073)
  • 0.010 -0.013 -0.040 0.445 0.047 Women Quota Bidding

  (0.035) (0.078) (0.078) (0.433) (0.073) 0.063 0.036 0.007 -0.270 0.125 No Women Quota (0.040) (0.088) (0.089) (0.490) (0.083)

  Appendix Table 4B: Raskin Distribution Process, by Women Quota Subtreatment Time to Distance to HH paid for Satisfied with

  Amount Satisfied with purchase point purchase point rice in Raskin Bought Raskin Price markup purchased rice quality (meters) (minutes) advance program (1) (2) (3) (4) (5) (6) (7) (8)

  Info or Bidding No Women Quota 0.029 48 0.701** 0.006 -10.893 0.682* 0.002 -0.011 (0.03) (34.63) (0.32) (0.01) (15.13) (0.35) (0.03) (0.01) Women Quota -0.047 -13 -0.403 0.006 9.003 0.193 0.024 -0.029** (0.03) (28.64) (0.26) (0.01) (22.45) (0.37) (0.03) (0.01) Bidding No Women Quota -0.027 -75* -0.362 0.024 17.642 -0.838** 0.005 -0.007 (0.04) (38.84) (0.37) (0.02) (18.24) (0.39) (0.04) (0.02) Women Quota 0.068* -20 0.363 0.014 -2.369 -0.159 -0.023 0.018 (0.04) (31.39) (0.29) (0.02) (24.49) (0.40) (0.03) (0.01) P-Value Info or Bidding: No Quota = Quota

  0.06

  0.15

  0.00

  0.97

  0.43

  0.31

  0.62

  0.32 Bidding: No Quota = Quota

  0.06

  0.28

  0.13

  0.65

  0.52

  0.22

  0.58

  0.25

  42 Observations 6,860 5,886 6,858 6,533 6,194 6,247 6,394 6,782 Control Mean 0.76 652.39

  5.76 0.51 190.96

  5.94

  0.43

  0.59 Note: This table replicates Table 3, but disaggregates bidding and information treatments by women quota subtreatment. Standard errors are clustered by village.

  • p<0.01, ** p<0.05, * p<0.1.

  

Appendix Table 5A: Who Distributes Raskin Six Months After Intervention, by Enhanced Supervision Subtreatment

Panel A: Distributor Identity and Connections

In charge of any Distributor/ spouse is Distributor/spouse responsibilities Distributor/spouse related to a local is/was or is related to Lives in before May 2013 is/was local official official hamlet official Is a trader locality Info or Bidding (1) (2) (3) (4) (5) (6)

No Enhanced -0.034 -0.120* 0.018 0.017 -0.031 -0.057

  

Supervision (0.064) (0.067) (0.059) (0.072) (0.035) (0.056)

Enhanced Supervision -0.035 0.001 0.067 -0.015 0.047 0.003

Bidding (0.064) (0.066) (0.059) (0.072) (0.035) (0.056)

No Enhanced -0.164** -0.038 0.073 -0.000 0.107*** 0.012

Supervision (0.073) (0.076) (0.067) (0.082) (0.040) (0.064)

Enhanced Supervision -0.166** -0.115 0.019 -0.004 0.035 -0.050

P-Value (0.072) (0.075) (0.067) (0.082) (0.040) (0.063) Info or Bidding: No

  0.987 0.167 0.531 0.733 0.093 0.415 Enhanced = Enhanced Bidding: No Enhanced 0.985 0.474 0.569 0.978 0.209 0.493 = Enhanced Control Mean

  0.803 0.347 0.160 0.350 0.034 0.816

Panel B: Other Distributor Characteristics

Has personal Owns a truck Avg Digit Span Raw dice score savings account and/or a boat above median above median Years of education for business Info or Bidding (1) (2) (3) (4) (5)

  • 0.052 -0.003 -0.007 0.104 0.064 No Enhanced Supervision (0.035) (0.078) (0.078) (0.432) (0.073) Enhanced Supervision -0.025 -0.014 -0.035 0.362 0.008 Bidding

  (0.035) (0.078) (0.078) (0.432) (0.073) No Enhanced 0.040 0.070 0.026 -0.357 0.024 Supervision (0.040) (0.088) (0.089) (0.492) (0.083)

  Appendix Table 5B: Raskin Distribution Process, by Enhanced Supervision Subtreatment Time to Distance to HH paid for Satisfied with

  Amount Satisfied with purchase point purchase point rice in Raskin Bought Raskin Price markup purchased rice quality (meters) (minutes) advance program (1) (2) (3) (4) (5) (6) (7) (8)

  Info or Bidding No Enhanced Supervision 0.026 82*** 0.269 0.007 6.913 0.899** 0.040 -0.008 (0.03) (31.86) (0.34) (0.01) (17.60) (0.40) (0.03) (0.01) Enhanced Supervision -0.044 -45 0.035 0.005 -8.599 -0.015 -0.015 -0.032** (0.03) (29.21) (0.26) (0.02) (20.45) (0.30) (0.03) (0.01) Bidding No Enhanced Supervision 0.015 -114*** 0.113 0.019 25.478 -0.951** -0.048 -0.008 (0.03) (34.44) (0.38) (0.02) (20.02) (0.43) (0.04) (0.02) Enhanced Supervision 0.026 16 -0.118 0.019 -10.087 -0.049 0.030 0.020 (0.04) (33.13) (0.30) (0.02) (22.23) (0.35) (0.04) (0.01) P-Value Info or Bidding:

  0.08

  0.00

  0.56

  

0.90

  0.54

  0.05

  0.20

  0.19 Not Enhanced = Enhanced Bidding:

  0.84

  0.01

  0.63

  

0.99

  0.23

  0.10

  0.13

  0.19

  44 Not Enhanced = Enhanced Observations 6,860 5,886 6,858 6,533 6,194 6,247 6,394 6,782 Control Mean

  0.76 652.39

  5.76 0.51 190.96

  5.94

  0.43

  0.59 Note: This table replicates Table 3, but disaggregates bidding and information treatments by enhanced supervision subtreatment. Standard errors are clustered by locality. *** p<0.01, ** p<0.05, * p<0.1.

  Appendix Table 6A: Randomization Checks for Bidding and Information Treatments Difference Between Treatment and Control, with Stratum FE

  Means Bidding Information Bidding versus versus versus N Control Control Information

  

Control Bidding Information

(1) (2) (3) (4) (5) (6) (7) Log per capita consumption 6859

  13.06

  13.04 13.05 -.02 -.01 -.01 (.02) (.02) (.02) Household head years of education 6787

  6.11

  6.16 5.97 .08 -.14 .22 (.13) (.16) (.17) Bought Raskin in last 2 months 6860 .72 .73 .77 .01 .05* -.04 (.02) (.02) (.03) Raskin price markup (Rp/kg) 5332 650.85 654.47 683.13 -16.82 20.12 -36.94 (22.79) (29.94) (31.09) Amount of Raskin purchased per month 6860

  5.27

  5.06 5.45 -.24 .12 -.36 (kg) (.23) (.26) (.28) Subsidy received (Rp./month) 6860 28141.77 26948.15 28952.66 -1375.06 450.69 -1825.75

  (1256.69) (1439.08) (1519.1) Overall satisfaction according to 6732

  0.57

  0.57 0.57 -.01 .01 household, scaled 0-1 (.01) (.01) (.01)

45 Exclusion error 4621 .2 .2 .16 -.01 -.04* .03

  (.02) (.02) (.03) Inclusion error 2239 0.57 .58 .62 .01 .06 -.04 (.03) (.04) (.04)

  Locality distance to subdistrict 572

  7.52

  5.86 7.72 -2.1 .45 -2.55 (1.36) (1.91) (1.93) Percentage of agriculture households in 572 .07 .07 .06 -.01 -.01* .01 RT

  (.01) (.01) (.01) Log locality size 572

  6.97

  6.91 7 -.04 .04 -.08 (.05) (.07) (.07) Locality head years of education 572

  13.39

  13.31 13.61 -.04 .23 -.27 (.18) (.21) (.23) Local government official is part of 572 0.68 .71 .63 .03 -.06 .08

  

Appendix Table 6B: Randomization Checks for Minimum Number of Bids Requirement

Difference Between Enhanced Means Comp. and Open Bidding, with

  N Regular Bids Enhanced Comp. Stratum FE (1) (2) (3) (4) Log per capita consumption 2292

  

13.05

13.04 -.02 (.02) Household head years of education 2268

  

6.13

6.19 .1 (.18) Bought Raskin in last 2 months 2292 .73 .73 .01 (.03) Raskin price markup (Rp/kg) 1779 674.97 634.35 -31.83 (31.48) Amount of Raskin purchased per month 2292

  

5.04

5.08 -.09 (kg) (.33) Subsidy received (Rp./month) 2292 26866.03 27029.42 -621.87

  (1779.63) Overall satisfaction according to 2243 .58 0.57 household, scaled 0-1

  (.01) Exclusion error 1535 .19 .2 .01 (.03) Inclusion error 757

  0.57 .6 .04 (.05) Locality distance to subdistrict 191

  

6.51

5.21 -1.6 (1.45) Percentage of agriculture households in 191 .07 .07 RT

  (.01) Log locality size 191

6.91

6.91 .02

  

Appendix Table 7: Raskin Distribution Process, Excluding Baseline Controls

Time to Distance to HH paid for Satisfied with

  Bought Amount Satisfied with purchase point purchase point rice in Raskin Raskin Price markup purchased rice quality (meters) (minutes) advance program (1) (2) (3) (4) (5) (6) (7) (8)

  

Panel A: Combined Effect of Outsourcing

Info or Bidding 0.008 23.090 0.203 0.006 16.143 0.582* -0.001 -0.021* (0.03) (29.38) (0.27) (0.01) (20.72) (0.30) (0.04) (0.01) Bidding 0.007 -56.772* -0.159 0.019* -1.946 -0.537* -0.008 0.007 (0.03) (30.43) (0.28) (0.01) (21.95) (0.31) (0.04) (0.01) P-Value Bidding = Ctl

  0.45

  0.13

  0.83

  

0.00

  0.35

  0.84

  0.75

  0.08 Observations 6,860 5,886 6,858 6,533 6,240 6,267 6,394 6,782 Control Mean 0.76 652.39

  5.76 0.51 193.05

  5.96

  0.43

  0.59 Panel B: Estimating Additional Effect of Competition Info or Bidding 0.008 23.148 0.203 0.006 16.190 0.582* -0.001 -0.021* (0.03) (29.36) (0.27) (0.01) (20.73) (0.30) (0.04) (0.01)

47 Regular Bids -0.001 -24.687 -0.171 0.016 24.185 -0.484 0.019 0.003

  (0.03) (36.39) (0.33) (0.01) (26.86) (0.37) (0.04) (0.01) Enhanced 0.015 -87.674*** -0.146 0.022 -27.509 -0.588* -0.035 0.012 Comp. (0.03) (32.66) (0.31) (0.01) (22.85) (0.35) (0.04) (0.01) P-Value Regular = Enh

  0.62

  0.06

  0.94

  

0.66

  0.03

  0.76

  0.21

  0.43 Regular = Ctl

  0.77

  0.96

  0.91

  

0.05

  0.06

  0.74

  0.59

  0.08 Enh. = Ctl

  0.37

  0.01

  0.82

  

0.01

  0.49

  0.98

  0.33

  0.31 Observations 6,860 5,886 6,858 6,533 6,240 6,267 6,394 6,782 Control Mean 0.76 652.39

  5.76 0.51 193.05

  5.96

  0.43

  0.59 Note: This table replicates Table 3, but excludes baseline controls. *** p<0.01, ** p<0.05, * p<0.1.

  Appendix Table 8A: Raskin Distribution Process, Eligible Households Only Distance to Time to HH paid for Satisfied with Bought Amount Satisfied with purchase point purchase point rice in Raskin program Raskin Price markup purchased rice quality (meters) (minutes) advance

  (1) (2) (3) (4) (5) (6) (7) (8) Panel A: Combined Effect of Outsourcing

  Info or Bidding -0.005 19.935 0.095 0.010 -11.826 0.382 0.001 -0.017 (0.02) (23.27) (0.27) (0.01) (16.50) (0.30) (0.02) (0.01) Bidding 0.012 -52.867** 0.080 0.018 17.804 -0.449 -0.001 0.003 (0.03) (23.86) (0.29) (0.01) (17.35) (0.31) (0.03) (0.01) P-Value Bidding = Ctl

  0.74

  0.06

  0.42

  0.00

  0.62

  0.76

  0.99

  0.11 Observations 4,621 4,327 4,619 4,565 4,391 4,434 4,549 4,609 Control Mean 0.84 642.03

  6.61 0.51 196.22

  6.17

  0.44

  0.60 Panel B: Estimating Additional Effect of Competition Info or Bidding -0.005 20.024 0.096 0.010 -11.701 0.382 0.001 -0.017 (0.02) (23.27) (0.27) (0.01) (16.51) (0.30) (0.02) (0.01) Regular Bids 0.015 -28.663 0.206 0.018 38.555* -0.399 0.029 0.002

  48 (0.03) (28.54) (0.35) (0.01) (20.94) (0.38) (0.03) (0.01) Enhanced 0.008 -76.828*** -0.047 0.018 -2.750 -0.499 -0.031 0.004

  Comp. (0.03) (25.76) (0.32) (0.01) (18.36) (0.34) (0.03) (0.01) P-Value Regular = Enh

  0.84

  0.07

  0.46

  0.97

  0.03

  0.77

  0.05

  0.90 Regular = Ctl

  0.69

  0.71

  0.31

  0.02

  0.10

  0.95

  0.20

  0.20 Enh. = Ctl

  0.90

  0.00

  0.85

  0.02

  0.30

  0.65

  0.23

  0.20 Observations 4,621 4,327 4,619 4,565 4,391 4,434 4,549 4,609 Control Mean 0.84 642.03

  6.61 0.51 196.22

  6.17

  0.44

  0.60 Note: This table replicates Table 3, but includes only households officially eligible for Raskin. *** p<0.01, ** p<0.05, * p<0.1.

  

Appendix Table 8B: Raskin Distribution Process, Ineligible Households Only

Distance to Time to HH paid for Satisfied with Bought Amount Satisfied with purchase point purchase point rice in Raskin program

Raskin Price markup purchased rice quality (meters) (minutes) advance

  (1) (2) (3) (4) (5) (6) (7) (8)

Panel A: Combined Effect of Outsourcing

  Info or Bidding -0.011 16.707 0.289 -0.002 24.742 0.611* 0.043 -0.024 (0.03) (36.61) (0.27) (0.01) (19.05) (0.35) (0.03) (0.01) Bidding 0.035 -36.949 -0.190 0.022 -14.818 -0.647* -0.029 0.010 (0.03) (38.57) (0.29) (0.02) (19.47) (0.36) (0.03) (0.01) P-Value Bidding = Ctl

  0.33

  0.42

  0.62

  

0.09

  0.49

  0.89

  0.55

  0.21 Observations 2,239 1,559 2,239 1,968 1,803 1,813 1,845 2,173 Control Mean 0.60 681.37

  4.01 0.50 178.02

  5.37

  0.39

  0.56 Panel B: Estimating Additional Effect of Competition Info or Bidding -0.011 16.604 0.289 -0.002 24.756 0.611* 0.043 -0.024 (0.03) (36.58) (0.27) (0.01) (19.05) (0.35) (0.03) (0.01)

49 Regular Bids 0.012 -11.879 -0.482 0.012 -6.228 -0.524 -0.038 -0.004

  (0.04) (47.15) (0.32) (0.02) (22.41) (0.40) (0.04) (0.02) Enhanced 0.059 -59.325 0.101 0.031* -22.852 -0.763* -0.021 0.023 Comp. (0.04) (40.00) (0.34) (0.02) (22.00) (0.41) (0.03) (0.02) P-Value Regular = Enh

  0.25

  0.24

  0.06

  

0.32

  0.44

  0.52

  0.60

  0.07 Regular = Ctl

  0.97

  0.90

  0.43

  

0.51

  0.30

  0.77

  0.88

  0.04 Enh. = Ctl

  0.12

  0.11

  0.14

  

0.06

  0.92

  0.64

  0.43

  0.99 Observations 2,239 1,559 2,239 1,968 1,803 1,813 1,845 2,173 Control Mean 0.60 681.37

  4.01 0.50 178.02

  5.37

  0.39

  0.56 Note: This table replicates Table 3, but includes only households officially ineligible for Raskin. *** p<0.01, ** p<0.05, * p<0.1. P-Value Mean Std. Dev. Mean Std. Dev. Losers = Winners

  (1) (2) (3) (4) (5) Price Markup (Rp/kg) Promised by Bidder 471.62 270.20 566.53 295.90 0.00*** Pay Before Receipt

  0.03

  0.35

  0.17

  0.38

  0.56 Raskin Distributed at Locality Level

  0.29

  0.46

  0.38

  0.49

  0.14 Raskin Distributed at Hamlet Level

  0.76

  0.43

  0.64 0.48 0.05** Raskin Distributed at Household Level

  0.03

  0.16

  0.17

  0.59 0.50 0.01** Pay After Receipt

  0.57 Bidder Offers Credit

  0.16

  0.37

  0.08 0.27 0.03** Price Markup (Rp/kg) Promised by Bidder 464.04 277.23 551.42 286.22 0.00*** Pay Before Receipt

  0.40

  0.49

  0.38

  0.49

  0.68 Pay During Receipt

  0.36

  0.48

  0.40

  0.49

  0.37 Appendix Table 9: Comparison of Winning and Losing Bids in Bidding Treatment Localities Losers Winners

  0.14

  0.50

  0.44

  0.15 Raskin Distributed at Hamlet Level

  0.50

  0.36 0.48 0.04** Pay During Receipt

  0.39

  0.49

  0.47 0.50 0.03** Pay After Receipt

  0.19

  0.40

  0.20

  0.40

  0.74 Raskin Distributed at Locality Level

  0.27

  0.44

  0.32

  0.47

  0.76

  0.43

  0.38

  0.34 0.48 0.02** Pay During Receipt

  0.50

  0.48

  0.43 Price Markup (Rp/kg) Promised by Bidder 479.30 264.45 588.22 309.44 0.00*** Pay Before Receipt

  0.36

  0.15

  0.17

  0.43

  0.37 Bidder Offers Credit

  0.20

  0.04

  0.17

  0.03

  0.68 0.47 0.01** Raskin Distributed at Household Level

  

Panel B: Bids in Regular Bidding Localities (178 bids in 79 localities)

Panel C: Bids in Localities with Bidding with Enhanced Competition (225 bids in 81 localities)

Panel A: Bids in All Bidding Localities (403 bids in 160 localities)

  

Appendix Table 10: Who was Selected in the Bidding Localities?

1-by-1 (Separate regressions) Joint (Form) Joint (All) (1) (2) (3)

  

Panel A: Proposal Characteristics

Price markup (Rp/kg) promised by bidder -0.010*** -0.009*** -0.007***

(0.003) (0.003) (0.003) Pay before receipt -0.606 -0.614 -2.473 (0.531) (1.170) (4.086) Offers credit 0.731 1.148 1.021 (0.599) (0.743) (0.816) Lives in distribution locality -1.207* -1.029 -0.792 (0.643) (1.054) (1.179) Is Raskin distributor at time of bidding 0.531* 0.297 0.344 (0.285) (0.346) (0.362)

Is a trader 1.120** 2.303*** 1.734***

(0.527) (0.714) (0.576) Has means of transportation supportive of Raskin 1.387*** 1.295*** 0.983 distribution (0.422) (0.494) (0.622)

  

Panel B: Individual Characteristics

Bidder/spouse is related to a local official 0.223 0.035 (0.317) (0.358) Bidder/spouse is/was local official 0.479 0.115 (0.454) (0.702) Raw dice score above median 0.197 0.105 (0.279) (0.332)

  

Appendix Table 11: Who was Selected in Bidding Localities, by Enhanced Comp.?

1-by-1 (Separate regressions) Joint (Form) Joint (All) (1) (2) (3)

  

Panel A: Proposal Characteristics

Enhanced Comp. * Promised price markup 0.002 0.005 -0.007 (0.006) (0.008) (0.013) Enhanced Comp. * Pay before receipt -1.188 -4.470* -62.377*** (1.338) (2.619) (5.065) Enhanced Comp. * Offers credit -2.016 -1.141 -33.758*** (1.281) (1.830) (2.989) Enhanced Comp. * Lives in locality 12.896*** 16.921*** 8.911*** (1.215) (2.801) (3.370) Enhanced Comp. * Distributor at time of bidding 0.830 1.006 1.191 (0.577) (1.023) (1.236) Enhanced Comp. * Trader 1.031 2.630* 20.489*** (1.054) (1.539) (2.061) Enhanced Comp. * Has means of transportation 2.882** 2.936** 31.886*** supportive of Raskin distribution (1.193) (1.434) (1.767)

  

Panel B: Individual Characteristics

Enhanced Comp. * Bidder/spouse is related to a 0.723 1.661 local official (0.654) (1.109) Enhanced Comp. * Bidder/spouse is/was local 1.503 6.112** official

  (0.984) (2.513) Enhanced Comp. * Raw dice score above median -0.255 -0.611 (0.568) (0.838)

  

Appendix Table 12A: Who was Selected in Bidding Localities (OLS)

1-by-1 Joint (Form) Joint (All) (1) (2) (3)

  

Panel A: Proposal Characteristics

Price markup (Rp/kg) promised by bidder -0.002*** -0.002*** -0.002*** (0.000) (0.000) (0.001) Pay before receipt

  • 0.195 -0.240 -0.530 (0.203) (0.251) (0.542) Offers credit 0.210 0.265 0.262 (0.205) (0.166) (0.339) Lives in distribution locality -0.387 -0.277 -0.257 (0.264) (0.319) (0.559)

  Is Raskin distributor at time of bidding 0.175 0.072 0.117 (0.115) (0.111) (0.174) Is a trader 0.371* 0.468*** 0.525** (0.188) (0.153) (0.239) Has means of transportation supportive of 0.373*** 0.250** 0.261 Raskin distribution (0.115) (0.119) (0.241)

  

Panel B: Individual Characteristics

Bidder/spouse is related to a local official 0.101 -0.023 (0.197) (0.169) Bidder/spouse is/was local official 0.200 -0.006 (0.262) (0.277) Raw dice score above median 0.088 0.018 (0.170) (0.140) Years of education 0.040* 0.004 (0.024) (0.025)

  

Appendix Table 12B: Who was Selected in Bidding Localities, by Enhanced Comp.? (OLS)

1-by-1 Joint (Form) Joint (All) (2) (3)

  (1) Panel A: Proposal Characteristics

  Enhanced Comp. * Promised price markup 0.000 0.001 0.001 (0.001) (0.001) (0.001) Enhanced Comp. * Pay before receipt -0.379 -0.823* -1.423** (0.534) (0.472) (0.686) Enhanced Comp. * Offers credit -0.494 -0.281 -0.802 (0.358) (0.323) (0.616) Enhanced Comp. * Lives in locality 0.676** 1.350*** 1.506** (0.277) (0.431) (0.740) Enhanced Comp. * Distributor at time of bidding 0.260 0.170 -0.049 (0.230) (0.234) (0.369) Enhanced Comp. * Trader 0.268 0.355 0.508 (0.417) (0.346) (0.503) Enhanced Comp. * Has means of transportation 0.477** 0.328 0.469 supportive of Raskin distribution (0.227) (0.250) (0.465)

  

Panel B: Individual Characteristics

Enhanced Comp. * Bidder/spouse is related to a 0.325 0.140 local official

  (0.393) (0.369) Enhanced Comp. * Bidder/spouse is/was local 0.602 0.455 official (0.494) (0.602) Enhanced Comp. * Raw dice score above median -0.109 -0.165

  (0.341) (0.295) Enhanced Comp. * Years of education 0.023 0.038 (0.050) (0.055) Enhanced Comp. * Has personal savings account 0.127 -0.010 (0.381) (0.401)

  

Appendix Table 14: When Did Original Distributor Win and Continue Distributing? (Unconditional)

Where Was No Bidding Meeting Held, or Meeting Had No Bids? Where Did Original Distributor Win? Where is Original Distributor Still Distributing?

  Joint Joint Joint (Household Joint (Form (Household Joint (Form (Household Joint (Form 1-by-1 Only) Only) Joint (All) 1-by-1 Only) Only) Joint (All) 1-by-1 Only) Only) Joint (All) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

  

Panel A: Reported by Households in Baseline

Avg Price Markup (Rp/kg) -0.00299*** -0.00291** -0.00337*** -0.00154*** -0.00148** -0.00199** -0.00088** -0.00074 -0.00093

(0.00089) (0.00115) (0.00122) (0.00054) (0.00063) (0.00097) (0.00044) (0.00049) (0.00069) HH Bought Raskin in Last 2 Months

  0.37

  0.52 0.29 1.47**

  1.03

  0.64

  0.22

  0.55

  0.74 (0.89) (1.24) (1.46) (0.60) (0.73) (0.93) (0.59) (0.69) (0.87)

Avg Amount of Raskin Purchased (kg) 0.0240 -0.0429 -0.0615 0.0618 0.0025 0.0298 -0.0304 -0.0300 -0.0588

(0.068) (0.091) (0.120) (0.070) (0.060) (0.063) (0.053) (0.057) (0.063) Avg Satisfaction with Program Quality 3.89*

  1.41 0.77 4.52*** 3.91** 5.57** -0.010 -1.25 -1.05 (0-1 scale) (2.24) (2.99) (3.37) (1.68) (1.83) (2.23) (1.55) (1.58) (1.86)

Avg Distance to Purchase Point (meters) 0.00078 -0.00070 -0.00030 -0.00121 -0.00200* -0.00223* 0.00013 0.00010 0.00019

  (0.0013) (0.00147) (0.00156) (0.00087) (0.00109) (0.00127) (0.00086) (0.00096) (0.00111) HH purchased Raskin in advance 1.47*** 1.10** 1.20** 1.07***

  0.45

  0.19

  0.44

  0.35

  0.51 (0.49) (0.53) (0.61) (0.36) (0.42) (0.50) (0.35) (0.40) (0.46)

55 Panel B: From Facilitation Forms

  

Raw dice score above median 0.81* 0.95** 1.01** 1.08*** 1.13*** 1.06** 0.65* 0.64* 0.65*

(0.46) (0.48) (0.49) (0.35) (0.36) (0.41) (0.33) (0.35) (0.38) Old Distributor Provides Credit if -0.69 -0.59 0.36 -0.37 -0.24 0.28 -0.29 -0.60 -0.51

Recipient Cannot Afford Raskin (0.65) (0.73) (0.80) (0.37) (0.53) (0.62) (0.37) (0.46) (0.48)

Costs of Rental Vehicle and/or Fuel to

  • 0.00620 -0.00655 -0.01197* -0.00016 -0.00370 -0.00121 0.00185 0.00199 0.00401 Old Distributor (0.00874) (0.00810) (0.00705) (0.00327) (0.00399) (0.00428) (0.00320) (0.00393) (0.00447)

    Non-Transportation Costs to Old -0.00031 0.00039 0.00253 -0.00113 -0.00054 0.00041 -0.00105 -0.00027 0.00041

    Distributor (0.00174) (0.00141) (0.00157) (0.00101) (0.00108) (0.00115) (0.00106) (0.00116) (0.00118)

    Joint P-Value 0.004 0.294 0.006 0.002 0.036 0.013 0.438 0.320 0.452 Observations 187 149 147 187 149 147 185 148 146

  Mean

  0.13

  0.17

  0.17

  0.59

  0.63

  0.63

  0.57

  0.58

  0.58

  Appendix Table 15: When Did Original Distributor Win and Continue Distributing? (OLS) Where Was No Bidding Meeting Held, or Meeting Where Did Original Distributor Win? Where is Original Distributor Still Distributing? Had No Bids? (Conditional on Bidding Held with 1+ Bid) (Conditional on Not Winning) Joint Joint Joint (Household Joint (Form (Household Joint (Form (Household Joint (Form

  1-by-1 Only) Only) Joint (All) 1-by-1 Only) Only) Joint (All) 1-by-1 Only) Only) Joint (All) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Panel A: Reported by Households in Baseline

  Avg Price Markup (Rp/kg) -0.00027*** -0.00021*** -0.00028*** -0.00027** -0.00024* -0.00027 0.00004 -0.00002 0.00016 (0.00007) (0.00008) (0.00011) (0.00013) (0.00014) (0.00018) (0.00017) (0.00020) (0.00031) HH Bought Raskin in Last 2 0.042 0.051 -0.004 0.383*** 0.215 0.154 -0.046 0.067 0.241 Months (0.097) (0.095) (0.145) (0.143) (0.167) (0.219) (0.196) (0.236) (0.326) Avg Amount of Raskin 0.0029 -0.0011 -0.0033 0.0146 -0.0013 0.0032 -0.0154 -0.0108 -0.0145 Purchased (kg) (0.0088) (0.0082) (0.0096) (0.0163) (0.0130) (0.0120) (0.0113) (0.0112) (0.0130) Avg Satisfaction with Program 0.454* 0.239 0.246 0.978** 0.905** 1.268*** 0.116 -0.040 0.074 Quality (0-1 scale) (0.265) (0.265) (0.344) (0.393) (0.433) (0.470) (0.506) (0.591) (0.916) Avg Distance to Purchase Point 0.00010 0.00003 0.00009 -0.00044* -0.00051** -0.00062** 0.00037 0.00040 0.00023 (meters) (0.00018) (0.00015) (0.00018) (0.00023) (0.00024) (0.00026) (0.00030) (0.00032) (0.00036) HH purchased Raskin in 0.167*** 0.109* 0.128* 0.197** 0.056 -0.018 -0.224* -0.194 -0.239 advance (0.057) (0.060) (0.070) (0.090) (0.098) (0.105) (0.117) (0.150) (0.229)

  Panel B: From Facilitation Forms

  56 Raw dice score above median 0.110* 0.131** 0.128** 0.238*** 0.236** 0.186** -0.048 -0.090 -0.115 (0.060) (0.064) (0.062) (0.087) (0.091) (0.092) (0.128) (0.136) (0.129) Old Distributor Provides Credit -0.068 -0.079 0.018 -0.064 -0.026 0.047 0.077 0.047 -0.107 if Recipient Cannot Afford (0.054) (0.086) (0.087) (0.097) (0.131) (0.137) (0.131) (0.189) (0.189) Costs of Rental Vehicle and/or -0.00058 -0.00085 -0.00088 0.00028 -0.00054 -0.00004 -0.00053 0.00074 0.00075 Fuel to Old Distributor (0.00065) (0.00084) (0.00067) (0.00089) (0.00110) (0.00108) (0.00150) (0.00172) (0.00187) Non-Transportation Costs to Old -0.00004 0.00007 0.00026 -0.00033 -0.00018 -0.00001 -0.00043 0.00007 -0.00011 Distributor (0.00019) (0.00024) (0.00025) (0.00029) (0.00033) (0.00030) (0.00034) (0.00062) (0.00070) Joint P-Value 0.002 0.268 0.012 0.002 0.113 0.000 0.442 0.964 0.362 Observations 187 149 147 162 123 122

  76

  55

  54 Mean

  0.13

  0.17

  0.17

  0.53

  0.55

  0.56

  0.32

  0.33

  0.31 Note: This table replicates Table 6, but estimates the coefficients by OLS rather than the logit.*** p<0.01, ** p<0.05, * p<0.1

  

Appendix Table 16: When Did Original Distributor Win and Continue Distributing? With Baseline Controls

Where Was No Bidding Meeting Held, or Meeting Where Did Original Distributor Win? Where is Original Distributor Still Distributing? Had No Bids? (Conditional on Bidding Held with 1+ Bid) (Conditional on Not Winning) Joint Joint Joint (Household Joint (Form (Household Joint (Form (Household Joint (Form

  1-by-1 Only) Only) Joint (All) 1-by-1 Only) Only) Joint (All) 1-by-1 Only) Only) Joint (All) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Panel A: Reported by Households in Baseline

Avg Price Markup (Rp/kg) -0.00336*** -0.00308** -0.00305** -0.00142** -0.00141** -0.00176* 0.00031 0.00001 0.00203

(0.00111) (0.00138) (0.00139) (0.00060) (0.00066) (0.00092) (0.00084) (0.00099) (0.00161) HH Bought Raskin in Last 2 0.234 0.974 1.105 1.601** 1.138 1.117 0.030 0.489 2.800

  

Months (1.068) (1.490) (1.793) (0.661) (0.831) (1.154) (0.924) (1.224) (2.237)

Avg Amount of Raskin 0.07536 0.00143 -0.00006 0.08532 0.01664 0.05495 -0.07359 -0.06960 -0.07413

Purchased (kg) (0.08219) (0.11415) (0.15475) (0.07315) (0.06628) (0.07213) (0.06319) (0.08769) (0.18843)

Avg Satisfaction with Program 5.193** 1.777 1.390 4.388** 3.494* 5.075** 0.814 0.388 -2.548

Quality (0-1 scale) (2.442) (2.996) (3.303) (1.831) (2.084) (2.341) (2.240) (2.721) (6.384)

Avg Distance to Purchase Point 0.00156 -0.00010 0.00052 -0.00178* -0.00249** -0.00333** 0.00145 0.00171 0.00174

(meters) (0.00129) (0.00152) (0.00177) (0.00104) (0.00126) (0.00148) (0.00138) (0.00146) (0.00214)

HH purchased Raskin in 1.580*** 1.290** 1.526** 0.868** 0.290 -0.094 -1.275** -1.190 -1.605

advance (0.553) (0.624) (0.684) (0.391) (0.441) (0.535) (0.647) (0.784) (1.658)

Panel B: From Facilitation Forms

57 Raw dice score above median 0.990** 1.121** 1.407** 1.084*** 1.176*** 1.082** 0.020 -0.156 -0.071

  (0.489) (0.495) (0.576) (0.382) (0.424) (0.458) (0.623) (0.716) (0.770)

Old Distributor Provides Credit -0.734 -0.681 -0.249 -0.270 -0.220 0.086 0.472 0.524 -0.753

if Recipient Cannot Afford (0.616) (0.731) (0.870) (0.400) (0.580) (0.652) (0.576) (1.010) (1.049)

Costs of Rental Vehicle and/or -0.00599 -0.00616 -0.00970 0.00024 -0.00330 -0.00008 -0.00150 -0.00002 -0.00345

Fuel to Old Distributor (0.00775) (0.00743) (0.00659) (0.00386) (0.00480) (0.00485) (0.00809) (0.00906) (0.00945)

Non-Transportation Costs to Old -0.00074 -0.00017 0.00131 -0.00174 -0.00104 -0.00018 -0.00172 0.00009 -0.00199

Distributor (0.00158) (0.00138) (0.00150) (0.00126) (0.00137) (0.00137) (0.00203) (0.00287) (0.00443)

Joint P-Value 0.003 0.149 0.005 0.012 0.071 0.020 0.526 0.991 0.092 Observations 187 149 147 162 123 122

  76

  55

  54 Mean

  0.13

  0.17

  0.17

  0.53

  0.55

  0.56

  0.32

  0.33

  0.31 Note: This table replicates Table 6, but controls for baseline characteristics: log locality population, number of hamlets in locality, and distance to subdistrict. *** p<0.01 ** p<0.05 * p<0.1

  Appendix Table 17: When Did Original Distributor Win and Continue Distributing? With Baseline Controls, Unconditional Where Was No Bidding Meeting Held, or Meeting Where is Original Distributor Still Had No Bids? Where Did Original Distributor Win? Distributing?

  Joint Joint Joint (Household Joint (Form (Household Joint (Form (Household Joint (Form 1-by-1 Only) Only) Joint (All) 1-by-1 Only) Only) Joint (All) 1-by-1 Only) Only) Joint (All) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

  Panel A: Reported by Households in Baseline Avg Price Markup (Rp/kg) -0.00336*** -0.00308** -0.00305** -0.00176*** -0.00169*** -0.00207** -0.001 -0.001 -0.000 (0.00111) (0.00138) (0.00139) (0.00057) (0.00063) (0.00087) (0.001) (0.001) (0.001) HH Bought Raskin in Last 2 0.234 0.974 1.105 1.427** 1.159 1.005 0.113 0.508 0.849 Months (1.068) (1.490) (1.793) (0.599) (0.764) (1.003) (0.596) (0.690) (0.848) Avg Amount of Raskin 0.07536 0.00143 -0.00006 0.0922 0.0303 0.0710 -0.023 -0.022 -0.014 Purchased (kg) (0.08219) (0.11415) (0.15475) (0.0710) (0.0650) (0.0706) (0.054) (0.060) (0.063) Avg Satisfaction with Program 5.193** 1.777 1.390 4.912*** 3.499* 4.720** 0.186 -1.121 -1.357 Quality (0-1 scale) (2.442) (2.996) (3.303) (1.744) (1.966) (2.229) (1.560) (1.604) (1.882) Avg Distance to Purchase Point 0.00156 -0.00010 0.00052 -0.00105 -0.0020* -0.00228* 0.000 0.000 0.001 (meters) (0.00129) (0.00152) (0.00177) (0.00090) (0.0011) (0.00130) (0.001) (0.001) (0.001) HH purchased Raskin in 1.580*** 1.290** 1.526** 1.085*** 0.517 0.269 0.353 0.343 0.623 advance (0.553) (0.624) (0.684) (0.376) (0.431) (0.530) (0.358) (0.404) (0.485)

  58 Panel B: From Facilitation Forms Raw dice score above median 0.990** 1.121** 1.407** 1.213*** 1.359*** 1.305*** 0.770** 0.767** 0.845** (0.489) (0.495) (0.576) (0.359) (0.406) (0.457) (0.351) (0.385) (0.423) Old Distributor Provides Credit -0.734 -0.681 -0.249 -0.386 -0.364 0.100 -0.215 -0.556 -0.678 if Recipient Cannot Afford (0.616) (0.731) (0.870) (0.386) (0.557) (0.630) (0.369) (0.463) (0.494) Costs of Rental Vehicle and/or -0.00599 -0.00616 -0.00970 -0.0008 -0.0047 -0.0018 0.00246 0.00167 0.00225 Fuel to Old Distributor (0.00775) (0.00743) (0.00659) (0.0037) (0.0044) (0.0043) (0.00341) (0.00431) (0.00472) Non-Transportation Costs to -0.00074 -0.00017 0.00131 -0.00158 -0.00096 -0.00003 -0.00098 -0.00033 -0.00016 Old Distributor (0.00158) (0.00138) (0.00150) (0.00103) (0.00115) (0.00128) (0.00103) (0.00120) (0.00127) Joint P-Value 0.003 0.149 0.005 0.001 0.016 0.004 0.760 0.244 0.682 Observations 187 149 147 187 149 147 185 148 146

  (1) (2) (3) (4) Info or Bidding 0.041 0.041 0.053 0.053 (0.049) (0.049) (0.049) (0.049) Bidding 0.160*** 0.128** (0.052) (0.052) Regular Bids 0.130** 0.108*

  (0.059) (0.060) Enhanced Comp. 0.191*** 0.148** (0.060) (0.060) Control Mean 0.193 0.193 0.193 0.193

  Bidding=Ctl 0.000 0.000 Regular=Min 0.309 0.501 Regular=Ctl 0.001 0.001 Min=Ctl 0.000 0.000

  New Distributor At Endline New Distributor After Re-Evaluation

  "New distributor at endline" is a binary variable that equals one if distributor at endline did not hold any pre-pilot distribution duties. "New distributor after re- evaluation" is a binary variable that equals one if distributor after the 6-month

evaluation meeting did not hold any pre-pilot distribution duties. For control localities

(as well as treatment localities that did not hold evaluation meetings), this is the same

as "New distributor at endline." Regressions estimated by OLS with strata fixed effects. Original localities (i.e. if locality split in two, only keep original part) only. Each column has 571 observations. *** p<0.01, ** p<0.05, * p<0.1

  

Appendix Table 18: Who Distributes Raskin, After Six Month Re-Evaluation

  Appendix Table 20: Price Gap in Bidding Localities Strata FE No FE

  (1) (2) Panel A: Winner no longer distributing (38 localities)

  • 42.677 -54.740 Winner Cannot Handle Distribution (68.506) (48.990)

  Panel B: All localities with winner (185 localities) Winner Cannot Handle Distribution -157.307* -44.595 (83.059) (60.251) Note: This table shows relationship between reason for default and the difference between endline and baseline price in bidding localities. Panel A consists of localities where the winner is no longer distributing for any reason, while Panel B is the full sample of bidding localities that chose a winner. Column 1 includes strata fixed effects, while Column 2 does not. *** p<0.01, ** p<0.05, * p<0.1

  60 Appendix Table 21: Bidding vs Info Meetings (Conditional on Meeting Occurring) Info Bidding Regular Bids Enhanced Comp. P-Value Bidding Open

  Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. = Info = Min Length of Meeting (Hrs)

  1.58

  0.50

  1.74

  

0.68

  1.69

  0.71

  1.79 0.64 0.04**

  0.34 Total Number of Meeting Attendees

  28.54

  14.78

  21.69

  

9.13

  21.09

  10.31

  22.31 7.75 0.00***

  0.36

  8.78

  8.95

  7.77

  7.63

  0.41 Number Raskin Beneficiaries Attending

  12.42

  13.96

  8.28

  

8.32

0.00*** Number Local Officials Attending

  12.35

  5.94

  9.42

  

5.83

  8.80

  5.95

  10.06 5.66 0.00***

  0.14 Any Questions/Comments Left at Meeting?

  0.92

  0.28

  0.88

  

0.33

  0.91

  0.29

  0.85

  0.36

  0.33

  0.18 Number of Questions/Comments

  4.15

  1.77

  4.60

  2.32

  0.17

  6.59

  5.10

  4.36

  

2.06

0.00*** Notes: This table provides summary statistics on bidding and information meetings. All data come from the forms that the facilitators used to document these meetings. We

  

Appendix Table 22: Bid Markups

Incumbent's Bid Average Bid Minimum Bid Endline Price - Winning Bid All Localities >1 Bidder All Localities >1 Bidder All Localities >1 Bidder All Localities >1 Bidder

  (1) (2) (3) (4) (5) (6) (7) (8) 10.435 3.360 20.664 11.940 -1.788 -0.996 -55.103 -81.285 Enhanced Comp.

  (46.351) (58.060) (32.303) (38.502) (32.012) (38.558) (41.190) (50.545) 0.563*** 0.556*** 0.591*** 0.587*** 0.566*** 0.560*** 0.169*** 0.170*** Baseline Markup (0.074) (0.088) (0.054) (0.062) (0.053) (0.060) (0.053) (0.063)

Constant 88.649* 108.682 84.695** 115.171** 57.352 60.637 93.074** 112.365**

(53.424) (68.749) (40.392) (48.990) (39.338) (47.031) (40.962) (47.733)

  146 110 165 115 165 115 118

  77 Observations 0.436 0.423 0.509 0.523 0.487 0.489 0.083 0.115 R-squared

Note: This table compares bid markups between minimum bids and Regular Bids treatment localities. Columns 1-2 report results on bids from distribution

incumbents (if they exist), Columns 3-4 report average bids, Columns 5-6 report minimum bid in a locality, and Columns 7-8 report the difference

between endline price and winning bid in a locality (if winning bidder is distributor at endline). Columns 1, 3, 5, and 7 include all localities with at least

one bid, while Columns 2, 4, 6, and 8 restrict to localities with multiple bids. Results from OLS with a dummy for missing baseline markup. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

  61 Appendix Table 23: Corruption on Locality Price Markup in Treatment Localities Baseline Raskin Markup

  Endline Raskin Markup (1) (2) (3) (4) (5) Raskin 60.317

  48.047 Distributor's (44.415) (42.449) Perception of 624.087***

  621.387*** Local Head's (168.818) (157.694) Perception of

  514.035** Raskin (220.860)

  

Appendix Table 24 : Price Difference

Endline Price- Endline Price Bid Price

  Enhanced Competition -44 -40 (27) (32)

  Observations 1,990 1,690 Regular Bids Mean 667.1 211.8 Note: In this table, we explore the relationship between assignment to enhanced competition treatment and endline price (col.1) controlling for baseline markup. For column 2, we explore the relationship between assignment to enhanced competition and markup difference. Both columns include strata fixed effects. Robust standard errors in parentheses.

  • p<0.01, ** p<0.05, * p<0.1

  

Appendix Figure 1: Distribution of Baseline Raskin Price Markup

  

Appendix Figure 2: Project Timeline

2013

  2014 Sep Mar Apr May Jun Jul Aug Oct Nov Dec Jan Feb Treatment Implementation

  Re-Evaluation Baseline Survey Endline Survey Note: This figure provides a timeline of the implementation and survey activities.

  

Appendix Figure 3: Why were you selected (or not) as winning distributor?

  63

  

Appendix Figure 4: Winning price markup, by baseline price markup

Note: This figure shows a scatterplot of baseline price markup in a locality (from baseline household

survey) vs price markup promised by the winning bidder (from bidding meeting form). A 45 degree

line is provided for reference; promised markup is higher than baseline for points above the line, and

lower for points below the line.

  

Should government service delivery be outsourced to the private sector? If so, how? We conduct

the first randomized field experiment on these issues, spread across 572 Indonesian localities.

We show that allowing for outsourcing the last mile of a subsidized food delivery program

reduced operating costs without sacrificing quality. However, prices paid by citizens were

lower only where we exogenously increased competition in the bidding process. Corrupt elites

attempted to block reform, but high rents in these areas also increased entry, offsetting this effect.

The results suggest that sufficient competition is needed to ensure citizens share the gains from

outsourcing.

Dokumen baru